Spaces:

DrAbbas
/

SONAR-AI

Running

App Files Files Community

SONAR-AI / app.py

DrAbbas

Update app.py

de55394 verified 1 day ago

raw

history blame contribute delete

140 kB

	#!/usr/bin/env python3
	"""
	╔══════════════════════════════════════════════════════════════╗
	║ 🔱 SONAR-AI v15.2 - YOLO11x + DSOS-BA Edition ║
	║ 8 AI Models + Detection + TSC HS Codes + Auth ║
	╠══════════════════════════════════════════════════════════════╣
	║ ✅ 7 Classification Models (Ensemble) ║
	║ ✅ YOLOv10x-cls (مدرّب سابقاً) ║
	║ ✅ YOLO11x-cls (92.6% Top-1, 97.4% Top-5) 🆕 ║
	║ ✅ 2,128 HS Codes from TSC Database ║
	║ ✅ أدوات معالجة صور السونار (12 فلتر) ║
	║ ✅ واجهة RTL عربية كاملة ║
	║ ✅ جدول إحصائيات مع قاعدة بيانات SQLite ║
	║ ✅ Login Authentication ║
	║ ✅ Real AI Inference (GPU) ║
	║ ✅ C# Desktop API Integration ║
	╚══════════════════════════════════════════════════════════════╝
	"""

	import gradio as gr
	import pandas as pd
	import numpy as np
	from datetime import datetime
	import random
	import os
	import json
	import glob
	from PIL import Image, ImageDraw, ImageFont
	import pickle
	from scipy import ndimage
	from scipy.signal import find_peaks
	from scipy.fft import fft2, fftshift
	import time as time_module
	import matplotlib
	matplotlib.use("Agg")
	import matplotlib.pyplot as plt
	import io, base64

	VERSION = "17.2"
	API_URL = "http://65.108.7.202:5555/api"

	# ═══════════════════════════════════════════════════════════════
	# 🔧 Detect environment
	# ═══════════════════════════════════════════════════════════════

	USE_GPU = False
	MODELS_LOADED = False

	try:
	import torch
	import torch.nn as nn
	from torchvision import transforms, models
	USE_GPU = torch.cuda.is_available()
	DEVICE = torch.device('cuda' if USE_GPU else 'cpu')
	print(f"✅ PyTorch loaded \| Device: {DEVICE}")
	except ImportError:
	print("⚠️ PyTorch not available - simulation mode")
	torch = None

	try:
	from ultralytics import YOLO
	YOLO_AVAILABLE = True
	print("✅ Ultralytics loaded")
	except ImportError:
	YOLO_AVAILABLE = False
	print("⚠️ Ultralytics not available")

	# ═══════════════════════════════════════════════════════════════
	# 🔐 تسجيل الدخول
	# ═══════════════════════════════════════════════════════════════

	USERS = {
	"admin": "sonar2026",
	"inspector": "inspect123",
	"customs": "customs456",
	"A": "1",
	"عباس": "1",
	}

	def authenticate(username, password):
	return username in USERS and USERS[username] == password

	# ═══════════════════════════════════════════════════════════════
	# 🏷️ الفئات (43 - matching Training.py)
	# ═══════════════════════════════════════════════════════════════

	CATEGORIES = [
	'appliances', 'auto_parts', 'bags', 'banana', 'batteries', 'beverages',
	'cables', 'canned_food', 'ceramic', 'chemicals', 'cleaning', 'clothes',
	'cooking_oil', 'cosmetics', 'electronics', 'fruits', 'furniture', 'glass',
	'kitchenware', 'lubricants', 'machinery', 'meat', 'medical', 'milk',
	'motorcycle', 'nuts', 'other', 'paper', 'pipes', 'plastic', 'rice',
	'seeds', 'shoes', 'snacks', 'spices', 'steel', 'sugar', 'tea', 'tires',
	'tools', 'toys', 'weapons', 'wood'
	]

	# ═══════════════════════════════════════════════════════════════
	# 🆕 Swin-V2 Categories
	# ═══════════════════════════════════════════════════════════════
	SWINV2_CATEGORIES = sorted([
	'appliances', 'auto_parts', 'bags', 'banana', 'batteries', 'beverages',
	'cables', 'canned_food', 'ceramic', 'chemicals', 'cleaning', 'clothes',
	'cooking_oil', 'cosmetics', 'electronics', 'fruits', 'furniture', 'glass',
	'kitchenware', 'lubricants', 'machinery', 'meat', 'medical', 'milk',
	'motorcycle', 'nuts', 'oil', 'paint', 'paper', 'perfume', 'pharmaceutical',
	'plastic', 'rice', 'rubber', 'shoes', 'spices', 'sugar', 'tea', 'textiles',
	'tires', 'tobacco', 'toys', 'weapons'
	])
	SWINV2_CONCEALMENT_CLASSES = ['match', 'no_match']
	SWINV2_RISK_CLASSES = ['critical', 'high', 'low', 'medium', 'safe']

	HS_CHAPTERS = {
	'02':'لحوم','04':'ألبان وبيض','08':'فواكه','09':'بن وشاي وبهارات','10':'حبوب',
	'12':'بذور','15':'دهون وزيوت','17':'سكر','19':'محضرات حبوب','20':'محضرات خضر وفواكه',
	'22':'مشروبات','27':'وقود معدني','33':'عطور ومستحضرات','34':'صابون ومنظفات',
	'36':'متفجرات','38':'منتجات كيماوية','39':'لدائن بلاستيك','40':'مطاط',
	'42':'مصنوعات جلدية','44':'خشب','48':'ورق وكرتون',
	'62':'ملابس','64':'أحذية','69':'منتجات خزفية','70':'زجاج',
	'72':'حديد وصلب','73':'مصنوعات حديد','82':'أدوات معدنية',
	'84':'آلات ومعدات','85':'أجهزة كهربائية','87':'سيارات ومركبات',
	'90':'أجهزة طبية','93':'أسلحة','94':'أثاث','95':'ألعاب','96':'مصنوعات متنوعة',
	}

	CARGO_DATABASE = {
	'furniture': {'ar':'أثاث','hs':'940360','duty':30,'ch':'94'},
	'steel': {'ar':'حديد','hs':'721049','duty':5,'ch':'72'},
	'paper': {'ar':'ورق','hs':'480519','duty':10,'ch':'48'},
	'clothes': {'ar':'ملابس','hs':'620342','duty':20,'ch':'62'},
	'other': {'ar':'أخرى','hs':'999999','duty':15,'ch':'99'},
	'machinery': {'ar':'آلات','hs':'847989','duty':5,'ch':'84'},
	'milk': {'ar':'منتجات حليب','hs':'040210','duty':10,'ch':'04'},
	'electronics': {'ar':'إلكترونيات','hs':'854370','duty':10,'ch':'85'},
	'auto_parts': {'ar':'قطع غيار سيارات','hs':'870899','duty':5,'ch':'87'},
	'appliances': {'ar':'أجهزة منزلية','hs':'851660','duty':20,'ch':'85'},
	'ceramic': {'ar':'سيراميك','hs':'691090','duty':15,'ch':'69'},
	'chemicals': {'ar':'كيميائيات','hs':'382499','duty':5,'ch':'38'},
	'plastic': {'ar':'بلاستيك','hs':'392690','duty':15,'ch':'39'},
	'banana': {'ar':'موز','hs':'080390','duty':5,'ch':'08'},
	'tires': {'ar':'إطارات','hs':'401110','duty':15,'ch':'40'},
	'tools': {'ar':'أدوات','hs':'820559','duty':10,'ch':'82'},
	'toys': {'ar':'ألعاب','hs':'950300','duty':20,'ch':'95'},
	'seeds': {'ar':'بذور','hs':'120991','duty':5,'ch':'12'},
	'tea': {'ar':'شاي','hs':'090230','duty':10,'ch':'09'},
	'cleaning': {'ar':'مواد تنظيف','hs':'340220','duty':10,'ch':'34'},
	'canned_food': {'ar':'أغذية معلبة','hs':'200899','duty':10,'ch':'20'},
	'nuts': {'ar':'مكسرات','hs':'080290','duty':10,'ch':'08'},
	'glass': {'ar':'زجاج','hs':'701090','duty':15,'ch':'70'},
	'cables': {'ar':'كابلات','hs':'854449','duty':10,'ch':'85'},
	'snacks': {'ar':'وجبات خفيفة','hs':'190590','duty':15,'ch':'19'},
	'rice': {'ar':'رز','hs':'100630','duty':10,'ch':'10'},
	'pipes': {'ar':'أنابيب','hs':'730890','duty':10,'ch':'73'},
	'cosmetics': {'ar':'مستحضرات تجميل','hs':'330499','duty':20,'ch':'33'},
	'meat': {'ar':'لحوم','hs':'020230','duty':5,'ch':'02'},
	'lubricants': {'ar':'زيوت تشحيم','hs':'271019','duty':5,'ch':'27'},
	'bags': {'ar':'حقائب','hs':'420222','duty':20,'ch':'42'},
	'cooking_oil': {'ar':'زيت طبخ','hs':'151190','duty':5,'ch':'15'},
	'beverages': {'ar':'مشروبات','hs':'220299','duty':15,'ch':'22'},
	'shoes': {'ar':'أحذية','hs':'640299','duty':20,'ch':'64'},
	'batteries': {'ar':'بطاريات','hs':'850760','duty':15,'ch':'85'},
	'wood': {'ar':'خشب','hs':'440799','duty':10,'ch':'44'},
	'fruits': {'ar':'فواكه','hs':'081090','duty':5,'ch':'08'},
	'motorcycle': {'ar':'دراجات نارية','hs':'871190','duty':20,'ch':'87'},
	'medical': {'ar':'مستلزمات طبية','hs':'901890','duty':0,'ch':'90'},
	'kitchenware': {'ar':'أدوات مطبخ','hs':'732393','duty':15,'ch':'73'},
	'spices': {'ar':'بهارات','hs':'090421','duty':10,'ch':'09'},
	'weapons': {'ar':'أسلحة','hs':'930100','duty':0,'ch':'93'},
	'sugar': {'ar':'سكر','hs':'170199','duty':5,'ch':'17'},
	}

	# ═══════════════════════════════════════════════════════════════
	# 📂 تحميل قاعدة TSC
	# ═══════════════════════════════════════════════════════════════

	TSC_DATABASE = {}

	def load_tsc_database():
	global TSC_DATABASE
	paths = [
	"نسخة_TSC_2025-12-09.xlsx", "نسخة TSC 2025-12-09.xlsx",
	"/app/نسخة_TSC_2025-12-09.xlsx", "/app/نسخة TSC 2025-12-09.xlsx",
	os.path.join(os.path.dirname(os.path.abspath(__file__)), "نسخة_TSC_2025-12-09.xlsx"),
	os.path.join(os.path.dirname(os.path.abspath(__file__)), "نسخة TSC 2025-12-09.xlsx"),
	]
	for pattern in ["/app/TSC.xlsx", "/app/نسخة.xlsx", "./TSC.xlsx", "./نسخة.xlsx"]:
	paths.extend(glob.glob(pattern))
	for path in paths:
	if os.path.exists(path):
	try:
	df = pd.read_excel(path, engine='openpyxl')
	for _, row in df.iterrows():
	hs = str(int(row['IDE_HSC_NB1']))
	if hs not in TSC_DATABASE:
	TSC_DATABASE[hs] = {
	'desc': str(row['GDS_DS2']).strip()[:100] if pd.notna(row['GDS_DS2']) else '',
	'avg_price': round(float(row['AVR_MNT']), 2) if pd.notna(row['AVR_MNT']) else 0,
	'unit': str(row['AVR_UNT']) if pd.notna(row['AVR_UNT']) else '',
	'nb5': str(row['IDE_HSC_NB5']) if pd.notna(row['IDE_HSC_NB5']) else '',
	'currency': 'USD',
	}
	print(f"✅ TSC loaded: {len(TSC_DATABASE)} HS codes from {path}")
	return
	except Exception as e:
	print(f"⚠️ Error loading {path}: {e}")
	print("⚠️ TSC file not found")

	load_tsc_database()

	# ═══════════════════════════════════════════════════════════════
	# 🤖 تحميل النماذج
	# ═══════════════════════════════════════════════════════════════

	CLASSIFICATION_MODELS = {}
	DETECTION_MODEL = None
	ENSEMBLE_CONFIG = None
	ANOMALY_MODELS = None
	risk_model = None
	SELECTED_FEATURES = None
	FEATURE_EXTRACTOR = None
	SWINV2_CLASSIFICATION = None
	SWINV2_CONCEALMENT = None
	SWINV2_RISK = None

	MODEL_REPO = "DrAbbas/SONAR-AI-Models"

	def find_model(name):
	# 1. Check local paths
	for base in [".", "/app", os.path.dirname(os.path.abspath(__file__))]:
	path = os.path.join(base, name)
	if os.path.exists(path):
	return path
	# 2. Download from HF Model repo
	try:
	from huggingface_hub import hf_hub_download
	path = hf_hub_download(repo_id=MODEL_REPO, filename=name)
	print(f"📥 Downloaded {name} from {MODEL_REPO}")
	return path
	except Exception as e:
	print(f"⚠️ Could not download {name}: {e}")
	return None

	def get_transforms():
	if torch is None:
	return None
	return transforms.Compose([
	transforms.Resize((224, 224)),
	transforms.ToTensor(),
	transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
	])

	def get_swinv2_transforms(img_size=256):
	"""Transforms for Swin-V2 trained models (256×256)"""
	if torch is None:
	return None
	return transforms.Compose([
	transforms.Resize((img_size, img_size)),
	transforms.ToTensor(),
	transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
	])

	if torch is not None:
	class SOSUFS(nn.Module):
	def __init__(self, inf, outf, k=16):
	super().__init__()
	self.centers = nn.Parameter(torch.randn(k, inf))
	self.importance = nn.Parameter(torch.ones(inf))
	self.fc = nn.Linear(inf, outf)
	self.bn = nn.BatchNorm1d(outf)
	def forward(self, x):
	w = x * torch.sigmoid(self.importance)
	d = torch.cdist(w.unsqueeze(1), self.centers.unsqueeze(0))
	s = torch.softmax(-d.squeeze(1), dim=-1)
	return nn.functional.gelu(self.bn(self.fc(w + s @ self.centers)))

	class DeepSOSUFS(nn.Module):
	def __init__(self, nc):
	super().__init__()
	self.backbone = models.efficientnet_b3(weights=None)
	f = self.backbone.classifier[1].in_features
	self.backbone.classifier = nn.Identity()
	self.attn = nn.Sequential(nn.Linear(f, f//4), nn.ReLU(), nn.Linear(f//4, f), nn.Sigmoid())
	self.s1 = SOSUFS(f, 512)
	self.s2 = SOSUFS(512, 256)
	self.fc = nn.Linear(256, nc)
	self.drop = nn.Dropout(0.3)
	def forward(self, x):
	x = self.backbone(x)
	x = x * self.attn(x)
	x = self.drop(self.s1(x))
	x = self.drop(self.s2(x))
	return self.fc(x)

	def load_all_models():
	global CLASSIFICATION_MODELS, DETECTION_MODEL, ENSEMBLE_CONFIG, MODELS_LOADED
	nc = len(CATEGORIES)
	if torch is None:
	print("⚠️ PyTorch not available")
	return

	# 1. ConvNeXt-V2
	path = find_model("convnext_best.pt")
	if path:
	try:
	import timm
	m = timm.create_model('convnext_base', pretrained=False, num_classes=nc)
	m.load_state_dict(torch.load(path, map_location=DEVICE))
	m.to(DEVICE).eval()
	CLASSIFICATION_MODELS['ConvNeXt-V2'] = m
	print(f"✅ ConvNeXt-V2 loaded")
	except Exception as e:
	print(f"⚠️ ConvNeXt: {e}")

	# 2. EfficientNet-V2
	path = find_model("efficientnet_best.pt")
	if path:
	try:
	import timm
	m = timm.create_model('tf_efficientnetv2_m', pretrained=False, num_classes=nc)
	checkpoint = torch.load(path, map_location=DEVICE)
	if isinstance(checkpoint, dict) and 'model_state_dict' in checkpoint:
	m.load_state_dict(checkpoint['model_state_dict'])
	else:
	m.load_state_dict(checkpoint)
	m.to(DEVICE).eval()
	CLASSIFICATION_MODELS['EfficientNet-V2'] = m
	print(f"✅ EfficientNet-V2 loaded")
	except Exception as e:
	print(f"⚠️ EfficientNet: {e}")

	# 3. ResNet152
	path = find_model("resnet152_best.pt")
	if path:
	try:
	m = models.resnet152(weights=None)
	m.fc = nn.Linear(m.fc.in_features, nc)
	checkpoint = torch.load(path, map_location=DEVICE)
	if isinstance(checkpoint, dict) and 'model_state_dict' in checkpoint:
	m.load_state_dict(checkpoint['model_state_dict'])
	else:
	m.load_state_dict(checkpoint)
	m.to(DEVICE).eval()
	CLASSIFICATION_MODELS['ResNet152'] = m
	print(f"✅ ResNet152 loaded")
	except Exception as e:
	print(f"⚠️ ResNet152: {e}")

	# 4. Deep-SOSUFS-v3
	path = find_model("sosufs_best.pt")
	if path:
	try:
	m = DeepSOSUFS(nc)
	checkpoint = torch.load(path, map_location=DEVICE)
	if isinstance(checkpoint, dict) and 'model_state_dict' in checkpoint:
	m.load_state_dict(checkpoint['model_state_dict'])
	else:
	m.load_state_dict(checkpoint)
	m.to(DEVICE).eval()
	CLASSIFICATION_MODELS['Deep-SOSUFS-v3'] = m
	print(f"✅ Deep-SOSUFS-v3 loaded")
	except Exception as e:
	print(f"⚠️ SOSUFS: {e}")

	# 5. YOLOv10x-cls
	path = find_model("yolov10x_cls_best.pt")
	if path and YOLO_AVAILABLE:
	try:
	CLASSIFICATION_MODELS['YOLOv10x-cls'] = YOLO(path)
	print(f"✅ YOLOv10x-cls loaded")
	except Exception as e:
	print(f"⚠️ YOLOv10x-cls: {e}")

	# 6. YOLO11x-cls (92.6% Top-1, 97.4% Top-5) 🆕
	path = find_model("yolo11x_cls_best.pt")
	if path and YOLO_AVAILABLE:
	try:
	CLASSIFICATION_MODELS['YOLO11x-cls'] = YOLO(path)
	print(f"✅ YOLO11x-cls loaded (92.6% Top-1)")
	except Exception as e:
	print(f"⚠️ YOLO11x-cls: {e}")

	# 7. Ensemble config
	path = find_model("ensemble_config.json")
	if path:
	try:
	with open(path) as f:
	ENSEMBLE_CONFIG = json.load(f)
	print(f"✅ Ensemble config loaded")
	except Exception as e:
	print(f"⚠️ Ensemble: {e}")

	# 8. Detection
	det_path = find_model("sonar_yolo11_detection_best.pt") or find_model("sonar_yolov8x_detection_best.pt") or find_model("best_detection.pt") or find_model("best.pt")
	if det_path and YOLO_AVAILABLE:
	try:
	DETECTION_MODEL = YOLO(det_path)
	print(f"✅ Detection loaded")
	except Exception as e:
	print(f"⚠️ Detection: {e}")

	# 9. Anomaly Detection
	global ANOMALY_MODELS, SELECTED_FEATURES, FEATURE_EXTRACTOR
	path = find_model("anomaly_models.pkl")
	if path:
	try:
	with open(path, 'rb') as f:
	ANOMALY_MODELS = pickle.load(f)
	print(f"✅ Anomaly loaded ({len(ANOMALY_MODELS)} models)")
	except Exception as e:
	print(f"⚠️ Anomaly: {e}")

	path = find_model("selected_features.npy")
	if path:
	try:
	SELECTED_FEATURES = np.load(path)
	print(f"✅ DSOS-BA features ({len(SELECTED_FEATURES)} selected)")
	except Exception as e:
	print(f"⚠️ Features: {e}")


	# Download risk calibration model
	risk_model = None
	try:
	risk_path = find_model("risk_calibration_model.pkl")
	if risk_path is None:
	risk_path = hf_hub_download(repo_id=MODEL_REPO, filename="risk_calibration_model.pkl", token=HF_TOKEN)
	with open(risk_path, 'rb') as f:
	risk_data = pickle.load(f)
	risk_model = risk_data['model']
	print("✅ Risk calibration model loaded")
	except Exception as e:
	risk_model = None
	print(f"⚠️ Risk model not loaded: {e}")

	if torch is not None:
	try:
	fe = models.efficientnet_v2_m(weights=None)
	fe.classifier = nn.Identity()
	fe.to(DEVICE).eval()
	FEATURE_EXTRACTOR = fe
	print(f"✅ Feature extractor ready")
	except Exception as e:
	print(f"⚠️ FE: {e}")

	# ═══════════════════════════════════════════════════════════════
	# 🆕 Swin-V2 Models
	# ═══════════════════════════════════════════════════════════════
	global SWINV2_CLASSIFICATION, SWINV2_CONCEALMENT, SWINV2_RISK

	# 10. Swin-V2 Classification (43 classes, 94.1% F1)
	swinv2_cls_path = find_model("classification/best_swinv2_43cls.pth")
	if swinv2_cls_path:
	try:
	import timm
	nc_swin = len(SWINV2_CATEGORIES)
	m = timm.create_model('swinv2_tiny_window8_256', pretrained=False, num_classes=nc_swin)
	ckpt = torch.load(swinv2_cls_path, map_location=DEVICE)
	if isinstance(ckpt, dict) and 'model_state_dict' in ckpt:
	m.load_state_dict(ckpt['model_state_dict'])
	else:
	m.load_state_dict(ckpt)
	m.to(DEVICE).eval()
	CLASSIFICATION_MODELS['Swin-V2'] = m
	SWINV2_CLASSIFICATION = m
	print(f"✅ Swin-V2 Classification loaded ({nc_swin}cls, 94.1% F1)")
	except Exception as e:
	print(f"⚠️ Swin-V2 Classification: {e}")

	# 11. Concealment Detection (98.9% F1)
	conc_path = find_model("concealment/best_eva02.pth") or find_model("concealment/best_swinv2.pth")
	if conc_path:
	try:
	import timm
	if 'eva02' in conc_path:
	m = timm.create_model('eva02_tiny_patch14_224', pretrained=False, num_classes=2)
	img_size_conc = 224
	else:
	m = timm.create_model('swinv2_tiny_window8_256', pretrained=False, num_classes=2)
	img_size_conc = 256
	ckpt = torch.load(conc_path, map_location=DEVICE)
	if isinstance(ckpt, dict) and 'model_state_dict' in ckpt:
	m.load_state_dict(ckpt['model_state_dict'])
	else:
	m.load_state_dict(ckpt)
	m.to(DEVICE).eval()
	m._conc_img_size = img_size_conc
	SWINV2_CONCEALMENT = m
	print(f"✅ Concealment loaded (98.9% F1)")
	except Exception as e:
	SWINV2_CONCEALMENT = None
	print(f"⚠️ Concealment: {e}")
	else:
	SWINV2_CONCEALMENT = None

	# 12. Risk Assessment (5 levels, 97.2% F1)
	risk_path = find_model("risk/best_swinv2_risk.pth")
	if risk_path:
	try:
	import timm
	m = timm.create_model('swinv2_tiny_window8_256', pretrained=False, num_classes=5)
	ckpt = torch.load(risk_path, map_location=DEVICE)
	if isinstance(ckpt, dict) and 'model_state_dict' in ckpt:
	m.load_state_dict(ckpt['model_state_dict'])
	else:
	m.load_state_dict(ckpt)
	m.to(DEVICE).eval()
	SWINV2_RISK = m
	print(f"✅ Swin-V2 Risk loaded (5 levels, 97.2% F1)")
	except Exception as e:
	SWINV2_RISK = None
	print(f"⚠️ Swin-V2 Risk: {e}")
	else:
	SWINV2_RISK = None

	MODELS_LOADED = len(CLASSIFICATION_MODELS) > 0
	anom_txt = f"anomaly({len(ANOMALY_MODELS)})" if ANOMALY_MODELS else "no-anomaly"
	conc_txt = "✅ Concealment" if SWINV2_CONCEALMENT else "❌ Concealment"
	risk_txt = "✅ Risk" if SWINV2_RISK else "❌ Risk"
	print(f"\n🔱 Total: {len(CLASSIFICATION_MODELS)} cls + {'1 det' if DETECTION_MODEL else '0 det'} + {anom_txt} + {conc_txt} + {risk_txt}")

	load_all_models()

	# ═══════════════════════════════════════════════════════════════
	# 🔬 Inference
	# ═══════════════════════════════════════════════════════════════


	def extract_features_from_image(img):
	if FEATURE_EXTRACTOR is None or torch is None:
	return None
	try:
	transform = get_transforms()
	if img.mode != 'RGB':
	img = img.convert('RGB')
	t = transform(img).unsqueeze(0).to(DEVICE)
	with torch.no_grad():
	return FEATURE_EXTRACTOR(t).cpu().numpy()[0]
	except:
	return None

	def check_anomaly(features):
	if ANOMALY_MODELS is None or features is None or SELECTED_FEATURES is None:
	return 0.0, False, {}
	try:
	feat_sel = features[SELECTED_FEATURES] if len(features) > len(SELECTED_FEATURES) else features
	x = feat_sel.reshape(1, -1)
	votes = 0
	details = {}
	for name, model in ANOMALY_MODELS.items():
	try:
	pred = model.predict(x)[0]
	is_anom = (pred == -1)
	votes += int(is_anom)
	details[name] = is_anom
	except:
	details[name] = False
	score = votes / max(len(ANOMALY_MODELS), 1)
	return score, votes >= 2, details
	except:
	return 0.0, False, {}

	def scan_container_regions(img):
	regions = []
	if img is None or ANOMALY_MODELS is None:
	return regions
	w_img, h_img = img.size
	for i in range(4):
	x = i * (w_img // 4)
	sw, sh = w_img // 4, h_img
	try:
	region = img.crop((x, 0, x+sw, sh))
	feat = extract_features_from_image(region)
	score, is_anom, det = check_anomaly(feat) if feat is not None else (0.0, False, {})
	except:
	score, is_anom, det = 0.0, False, {}
	pos = ["مقدمة","وسط-1","وسط-2","نهاية"][i]
	status = 'danger' if score >= 0.67 else ('warning' if is_anom else 'normal')
	label = f"⚠️ مشبوه - {pos}" if is_anom else f"✅ طبيعي - {pos}"
	regions.append({'bbox':(x,0,sw,sh),'anomaly_score':score,'is_anomaly':is_anom,
	'status':status,'label_ar':label,'section':i+1})
	return regions

	def draw_arabic_boxes(img, regions):
	if not regions:
	return img
	draw = ImageDraw.Draw(img)
	try:
	font = ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf", 14)
	except:
	font = ImageFont.load_default()
	for r in regions:
	x, y, w, h = r['bbox']
	color = (255,0,0) if r['status']=='danger' else ((255,165,0) if r['status']=='warning' else (0,200,0))
	thick = 4 if r['status']=='danger' else 2
	for t in range(thick):
	draw.rectangle([x-t, y-t, x+w+t, y+h+t], outline=color)
	txt = f"{r['label_ar']} {r['anomaly_score']:.0%}"
	draw.rectangle([x, y, x+len(txt)*7+10, y+20], fill=color)
	draw.text((x+5, y+2), txt, fill=(255,255,255), font=font)
	return img

	def classify_single_region(model, region_img, categories, img_size=256):
	"""تصنيف منطقة واحدة بنموذج Swin-V2"""
	if torch is None or model is None:
	return None, 0.0
	try:
	tf = get_swinv2_transforms(img_size)
	inp = tf(region_img.convert('RGB')).unsqueeze(0).to(DEVICE)
	with torch.no_grad():
	out = model(inp)
	probs = torch.softmax(out, dim=1)[0].cpu().numpy()
	top_idx = int(probs.argmax())
	if top_idx < len(categories):
	return categories[top_idx], float(probs[top_idx])
	except Exception as e:
	print(f"⚠️ Region classify: {e}")
	return None, 0.0

	def classify_regions(img, n_regions=4, min_confidence=0.30):
	"""🆕 تقسيم صورة الحاوية إلى مناطق وتصنيف كل منطقة بـ Swin-V2"""
	if img is None or SWINV2_CLASSIFICATION is None:
	return []
	w, h = img.size
	region_w = w // n_regions
	results = []
	pos_names = ["المقدمة", "وسط-1", "وسط-2", "النهاية"] if n_regions == 4 else [f"قسم-{i+1}" for i in range(n_regions)]
	for i in range(n_regions):
	x1 = i * region_w
	x2 = min(x1 + region_w, w)
	region = img.crop((x1, 0, x2, h))
	cat, conf = classify_single_region(SWINV2_CLASSIFICATION, region, SWINV2_CATEGORIES, img_size=256)
	if cat and conf >= min_confidence:
	info = get_hs_info(cat)
	results.append({'region': i+1, 'position': pos_names[i] if i < len(pos_names) else f"قسم-{i+1}",
	'category': cat, 'category_ar': info['ar'], 'hs_code': info['hs'], 'duty': info['duty'],
	'confidence': conf, 'bbox': (x1, 0, x2-x1, h)})
	return results

	def draw_region_boxes(img, region_results):
	"""رسم مربعات التصنيف على أقسام الحاوية"""
	if not region_results: return img
	draw = ImageDraw.Draw(img)
	try:
	font_sm = ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf", 12)
	except:
	font_sm = ImageFont.load_default()
	colors = [(46,125,50),(21,101,192),(106,27,154),(230,81,0),(198,40,40),(0,131,143)]
	unique_cats = list(set(r['category'] for r in region_results))
	for r in region_results:
	x, y, w, h = r['bbox']
	color = colors[unique_cats.index(r['category']) % len(colors)]
	for t in range(3): draw.rectangle([x+t, y+t, x+w-t, y+h-t], outline=color)
	label = f"{r['position']}: {r['category_ar']} ({r['confidence']:.0%})"
	draw.rectangle([x, y, x+len(label)*8+10, y+24], fill=color)
	draw.text((x+5, y+3), label, fill=(255,255,255), font=font_sm)
	if len(unique_cats) > 1:
	alert = f"⚠️ {len(unique_cats)} أنواع بضائع مختلفة!"
	img_w = img.size[0]
	draw.rectangle([img_w//2-120, 2, img_w//2+120, 28], fill=(198,40,40))
	draw.text((img_w//2-110, 5), alert, fill=(255,255,255), font=font_sm)
	return img

	def classify_image(img):
	if not MODELS_LOADED or img is None:
	return simulate_classification()

	transform_224 = get_transforms()
	transform_256 = get_swinv2_transforms(256)

	if img.mode != 'RGB':
	img = img.convert('RGB')

	input_224 = transform_224(img).unsqueeze(0).to(DEVICE)
	input_256 = transform_256(img).unsqueeze(0).to(DEVICE) if transform_256 else input_224

	all_preds = {}
	model_results = {}

	for name, model in CLASSIFICATION_MODELS.items():
	try:
	if name in ('YOLOv10x-cls', 'YOLO11x-cls'):
	res = model(img, verbose=False)
	if res and res[0].probs is not None:
	probs = res[0].probs.data.cpu().numpy()
	for idx in range(min(len(probs), len(CATEGORIES))):
	cat = CATEGORIES[idx]
	all_preds.setdefault(cat, []).append(float(probs[idx]))
	top_idx = probs.argmax()
	if top_idx < len(CATEGORIES):
	model_results[name] = {'top1': CATEGORIES[top_idx], 'confidence': float(probs[top_idx])}
	continue

	elif name == 'Swin-V2':
	input_tensor = input_256
	categories = SWINV2_CATEGORIES
	else:
	input_tensor = input_224
	categories = CATEGORIES

	with torch.no_grad():
	out = model(input_tensor)
	probs = torch.softmax(out, dim=1)[0].cpu().numpy()
	for idx in range(min(len(probs), len(categories))):
	cat = categories[idx]
	all_preds.setdefault(cat, []).append(float(probs[idx]))
	top_idx = probs.argmax()
	if top_idx < len(categories):
	model_results[name] = {'top1': categories[top_idx], 'confidence': float(probs[top_idx])}
	except Exception as e:
	print(f"⚠️ {name}: {e}")

	ensemble = {}
	for cat, scores in all_preds.items():
	avg = np.mean(scores)
	if avg > 0.05:
	ensemble[cat] = avg

	sorted_results = sorted(ensemble.items(), key=lambda x: x[1], reverse=True)[:5]
	return {'ensemble': sorted_results, 'models': model_results, 'n_models': len(model_results)}

	def detect_objects(img):
	if DETECTION_MODEL is None or img is None:
	return None, []
	try:
	results = DETECTION_MODEL(img, conf=0.25, verbose=False)
	if results and len(results) > 0:
	r = results[0]
	detections = []
	if r.boxes is not None:
	for box in r.boxes:
	detections.append({
	'name': r.names.get(int(box.cls[0]), 'unknown'),
	'confidence': float(box.conf[0]),
	})
	annotated = Image.fromarray(r.plot()[..., ::-1])
	return annotated, detections
	except Exception as e:
	print(f"⚠️ Detection: {e}")
	return None, []

	def simulate_classification():
	n = random.randint(3, 6)
	items = random.sample(list(CARGO_DATABASE.keys()), n)
	ensemble = sorted([(it, random.uniform(0.6, 0.99)) for it in items], key=lambda x: x[1], reverse=True)
	return {'ensemble': ensemble, 'models': {'Simulation': {'top1': items[0], 'confidence': 0.95}}, 'n_models': 0}

	# ═══════════════════════════════════════════════════════════════
	# 🔗 HS Lookup
	# ═══════════════════════════════════════════════════════════════

	def lookup_hs(hs_code):
	hs = str(hs_code).strip()
	if hs in TSC_DATABASE:
	return TSC_DATABASE[hs]
	for length in [6, 4]:
	matches = {k: v for k, v in TSC_DATABASE.items() if k.startswith(hs[:length])}
	if matches:
	return list(matches.values())[0]
	return None

	def get_hs_info(cargo_key):
	if cargo_key not in CARGO_DATABASE:
	return {'en': cargo_key.upper(), 'ar': cargo_key, 'hs': '999999', 'ch': '99',
	'ch_name': '', 'duty': 15, 'tsc_desc': '', 'avg_price': 0, 'unit': '', 'tsc_code': ''}
	c = CARGO_DATABASE[cargo_key]
	tsc = lookup_hs(c['hs'])
	return {
	'en': cargo_key.upper().replace('_', ' '), 'ar': c['ar'], 'hs': c['hs'], 'ch': c['ch'],
	'ch_name': HS_CHAPTERS.get(c['ch'], ''), 'duty': c['duty'],
	'tsc_desc': tsc['desc'] if tsc else c['ar'],
	'avg_price': tsc['avg_price'] if tsc else 0,
	'unit': tsc['unit'] if tsc else '',
	'tsc_code': tsc['nb5'] if tsc else '',
	}

	# ═══════════════════════════════════════════════════════════════
	# 📊 البيانات
	# ═══════════════════════════════════════════════════════════════

	def get_stats():
	return {'total': 6707, 'match': 6034, 'mismatch': 673, 'high_risk': 294,
	'hs_codes': len(TSC_DATABASE), 'cargo_types': len(CARGO_DATABASE),
	'models': len(CLASSIFICATION_MODELS), 'detection': DETECTION_MODEL is not None}

	def get_anomalies():
	rows = []
	for i in range(30):
	items = random.sample(list(CARGO_DATABASE.keys()), random.randint(1, 4))
	rows.append({
	'الرقم': 6700 - i,
	'الحاوية': f'TCNU{random.randint(1000000,9999999)}',
	'الأصناف': ' + '.join([CARGO_DATABASE[it]['ar'] for it in items]),
	'أكواد_HS': ' \| '.join([CARGO_DATABASE[it]['hs'] for it in items]),
	'الخطورة': random.randint(0, 5),
	'التاريخ': datetime.now().strftime('%Y-%m-%d'),
	})
	return pd.DataFrame(rows)

	def get_categories():
	rows = []
	for i, (en, c) in enumerate(CARGO_DATABASE.items()):
	tsc = lookup_hs(c['hs'])
	rows.append({
	'#': i+1, 'EN': en.upper().replace('_',' '), 'AR': c['ar'], 'HS': c['hs'],
	'الفصل': f"{c['ch']}-{HS_CHAPTERS.get(c['ch'],'')}",
	'الرسوم%': c['duty'],
	'السعر': f"${tsc['avg_price']}" if tsc and tsc['avg_price'] > 0 else '—',
	})
	return pd.DataFrame(rows)

	def search_tsc(query):
	if not query or len(query.strip()) < 2:
	return pd.DataFrame([{'ملاحظة': 'اكتب كود HS أو وصف (حرفين على الأقل)'}])
	q = query.strip()
	results = []
	for hs, info in TSC_DATABASE.items():
	if q in hs or q in info['desc']:
	results.append({'كود_HS': hs, 'الرمز': info['nb5'], 'الوصف': info['desc'],
	'السعر': f"${info['avg_price']}" if info['avg_price'] > 0 else '—', 'الوحدة': info['unit']})
	if len(results) >= 50:
	break
	return pd.DataFrame(results) if results else pd.DataFrame([{'نتيجة': f'لا توجد نتائج لـ: {q}'}])

	# ═══════════════════════════════════════════════════════════════
	# 🔬 التحليل الشامل
	# ═══════════════════════════════════════════════════════════════


	# ═══════════════════════════════════════════════════════════════
	# 🎛️ شريط أدوات السونار (Web Version)
	# ═══════════════════════════════════════════════════════════════

	def apply_grayscale(img):
	if img is None: return None
	import numpy as np
	arr = np.array(img.convert('RGB'))
	gray = np.dot(arr[...,:3], [0.299, 0.587, 0.114]).astype(np.uint8)
	return Image.fromarray(np.stack([gray]*3, axis=-1))

	def apply_invert(img):
	if img is None: return None
	import numpy as np
	arr = np.array(img.convert('RGB'))
	return Image.fromarray(255 - arr)

	def apply_thermal(img):
	if img is None: return None
	try:
	import cv2
	except: return img
	import numpy as np
	arr = np.array(img.convert('RGB'))
	gray = cv2.cvtColor(arr, cv2.COLOR_RGB2GRAY)
	thermal = cv2.applyColorMap(gray, cv2.COLORMAP_JET)
	return Image.fromarray(cv2.cvtColor(thermal, cv2.COLOR_BGR2RGB))

	def apply_cool(img):
	if img is None: return None
	try:
	import cv2
	except: return img
	import numpy as np
	arr = np.array(img.convert('RGB'))
	gray = cv2.cvtColor(arr, cv2.COLOR_RGB2GRAY)
	cool = cv2.applyColorMap(gray, cv2.COLORMAP_WINTER)
	return Image.fromarray(cv2.cvtColor(cool, cv2.COLOR_BGR2RGB))

	def apply_rainbow(img):
	if img is None: return None
	try:
	import cv2
	except: return img
	import numpy as np
	arr = np.array(img.convert('RGB'))
	gray = cv2.cvtColor(arr, cv2.COLOR_RGB2GRAY)
	rainbow = cv2.applyColorMap(gray, cv2.COLORMAP_RAINBOW)
	return Image.fromarray(cv2.cvtColor(rainbow, cv2.COLOR_BGR2RGB))

	def apply_edge(img):
	if img is None: return None
	try:
	import cv2
	except: return img
	import numpy as np
	arr = np.array(img.convert('RGB'))
	gray = cv2.cvtColor(arr, cv2.COLOR_RGB2GRAY)
	edges = cv2.Canny(gray, 50, 150)
	colored = cv2.addWeighted(arr, 0.7, cv2.cvtColor(edges, cv2.COLOR_GRAY2RGB), 0.3, 0)
	return Image.fromarray(colored)

	def apply_contrast(img):
	if img is None: return None
	try:
	import cv2
	except: return img
	import numpy as np
	arr = np.array(img.convert('RGB'))
	lab = cv2.cvtColor(arr, cv2.COLOR_RGB2LAB)
	l, a, b = cv2.split(lab)
	clahe = cv2.createCLAHE(clipLimit=3.0, tileGridSize=(8,8))
	l = clahe.apply(l)
	enhanced = cv2.merge([l, a, b])
	return Image.fromarray(cv2.cvtColor(enhanced, cv2.COLOR_LAB2RGB))

	def apply_brightness(img):
	if img is None: return None
	import numpy as np
	arr = np.array(img.convert('RGB')).astype(np.int16)
	arr = np.clip(arr + 40, 0, 255).astype(np.uint8)
	return Image.fromarray(arr)


	# ═══════════════════════════════════════════════════════════════
	# 📡 ربط قاعدة البيانات عبر API
	# ═══════════════════════════════════════════════════════════════

	# ═══════════════════════════════════════════════════════════════
	# 🔬🌊 8 Novel Physics Techniques (C11-C18) — 244D
	# Dr. Abbas Fadel Jassim Al-Jubouri \| UKM \| 2026
	# ═══════════════════════════════════════════════════════════════

	SMUGGLING_PATTERNS_P = {
	'drugs_in_dates':{'density_range':(0.35,0.55),'texture':'granular'},
	'pills_in_candy':{'density_range':(0.30,0.50),'texture':'uniform'},
	'powder_in_tea':{'density_range':(0.25,0.45),'texture':'fine'},
	'liquid_in_oil':{'density_range':(0.20,0.40),'texture':'smooth'},
	'metal_in_parts':{'density_range':(0.60,0.90),'texture':'sharp'},
	'plastic_in_toys':{'density_range':(0.15,0.35),'texture':'mixed'},
	'cigs_in_fabric':{'density_range':(0.25,0.50),'texture':'layered'},
	'cash_in_books':{'density_range':(0.40,0.65),'texture':'stacked'},
	}

	def physics_preprocess(image, size=224):
	if image is None: return None
	if isinstance(image, np.ndarray): img = Image.fromarray(image)
	else: img = image
	return np.array(img.convert('L').resize((size,size)), dtype=np.float32)/255.0

	def extract_dpm(img):
	H,W = img.shape; mh,mw = H//2,W//2
	quads = [img[:mh,:mw],img[:mh,mw:],img[mh:,:mw],img[mh:,mw:]]
	means = [q.mean() for q in quads]; stds = [q.std() for q in quads]; gm = img.mean()
	return np.array([max(means)-min(means),max(stds)-min(stds),np.std(means),sum(abs(m-gm) for m in means)/4], dtype=np.float32)

	def extract_ckb(img):
	features = []; md,sd = img.mean(),img.std()
	gm = np.sqrt(ndimage.sobel(img,0)2+ndimage.sobel(img,1)2).mean()
	for _,pat in SMUGGLING_PATTERNS_P.items():
	lo,hi = pat['density_range']; dm = min(max(0,1.0-abs(md-(lo+hi)/2)/((hi-lo)/2+1e-8)),1.0)
	t=pat['texture']; ts=0
	if t=='granular': ts=min(sd/0.15,1)
	elif t=='uniform': ts=max(0,1-sd/0.10)
	elif t=='fine': ts=min(sd/0.12,1)*0.8
	elif t=='smooth': ts=max(0,1-gm/0.15)
	elif t=='sharp': ts=min(gm/0.20,1)
	elif t=='mixed': ts=min(sdgm10,1)
	elif t=='layered': ts=min(abs(np.diff(img.mean(axis=1))).mean()/0.02,1)
	elif t=='stacked': ts=min(abs(np.diff(img.mean(axis=0))).mean()/0.02,1)
	features.append(0.6dm+0.4ts)
	return np.array(features, dtype=np.float32)

	def extract_3dtw(img):
	H,W = img.shape; surface = ndimage.gaussian_filter(img,1.5)
	gx,gy = ndimage.sobel(surface,1),ndimage.sobel(surface,0); grad_mag = np.sqrt(gx2+gy2)
	features = [surface.mean(),surface.std(),surface.max(),surface.min(),np.percentile(surface,75)-np.percentile(surface,25)]
	features += [grad_mag.mean(),grad_mag.std(),grad_mag.max(),np.percentile(grad_mag,90),(grad_mag>grad_mag.mean()+1.5grad_mag.std()).sum()/(HW)]
	xx,yy = np.meshgrid(np.linspace(0,2np.pi,W),np.linspace(0,2np.pi,H))
	for f in [2,4,8,16,32]:
	wave = surfacenp.sin(fxx)np.sin(fyy); power = np.abs(fftshift(fft2(wave)))**2
	features += [power.sum()/(H*W),power.max()]
	flat = surface.flatten(); mu,sigma = flat.mean(),flat.std()
	peaks,props = find_peaks(flat,height=mu+1.5*sigma,prominence=0.1)
	features += [len(peaks)/len(flat),props['prominences'].mean() if len(peaks)>0 else 0,ndimage.laplace(surface).std()]
	edges = (grad_mag>grad_mag.mean()+grad_mag.std()).astype(float)
	features += [edges.mean(),ndimage.label(edges)[1]/(H*W/100)]
	return np.array(features[:25], dtype=np.float32)

	def extract_cws(img):
	H,W = img.shape; cy,cx = H//2,W//2; y,x = np.ogrid[:H,:W]
	dist = np.sqrt((x-cx)2+(y-cy)2); max_r = np.sqrt(cx2+cy2); features = []
	for i in range(8):
	mask = (dist>=imax_r/8)&(dist<(i+1)max_r/8)
	rv = img[mask] if mask.sum()>0 else np.array([0.0]); features += [rv.mean(),rv.std()]
	for d in [img[:,:W//2],img[:,W//2:],img[:H//2,:],img[H//2:,:]]:
	gm = np.sqrt(ndimage.sobel(d,1)2+ndimage.sobel(d,0)2); features += [d.mean(),d.std(),gm.mean(),gm.max()]
	for s in [3,1.5,0.5,0]:
	sm = ndimage.gaussian_filter(img,s) if s>0 else img; e = np.sqrt(ndimage.sobel(sm,0)2+ndimage.sobel(sm,1)2)
	features += [e.mean(),e.std(),e.max(),(e>e.mean()+2e.std()).sum()/(HW)]
	rm = features[:16:2]; features += [np.std(rm),max(rm)-min(rm)]
	return np.array(features[:50], dtype=np.float32)

	def extract_tcv(img):
	H,W = img.shape; features = []
	for cy_c,cx_c in [(H//2,W//4),(H//2,W//2),(H//2,3*W//4)]:
	radius = min(H,W)//4; t = np.linspace(0,62np.pi,300); r = radius(1-t/(62*np.pi))
	px = (cx_c+rnp.cos(t)).astype(int); py = (cy_c+rnp.sin(t)).astype(int)
	v = (px>=0)&(px<W)&(py>=0)&(py<H); px,py = px[v],py[v]
	sampled = img[py,px] if len(px)>0 else np.array([0.0])
	c_feats = [sampled.mean(),sampled.std(),sampled.max()-sampled.min(),np.abs(np.diff(sampled)).mean() if len(sampled)>1 else 0]
	for d in range(5):
	sm = ndimage.gaussian_filter(img,0.5+d*1.0); dv = sm[py,px] if len(px)>0 else np.array([0.0]); c_feats += [dv.mean(),dv.std()]
	features += c_feats
	while len(features)<42: features.append(0.0)
	return np.array(features[:42], dtype=np.float32)

	def extract_dwe(img):
	H,W = img.shape; features = []; dark_mask = img<0.25; black_mask = img<0.10
	dv = img[dark_mask] if dark_mask.sum()>0 else np.array([0.0])
	features += [dark_mask.sum()/(HW),black_mask.sum()/(HW),dv.mean(),dv.std(),img.mean()]
	enhanced = np.clip(img*(1.0/(img.mean()+0.01)),0,1)
	de = enhanced[dark_mask] if dark_mask.sum()>0 else np.array([0.0])
	features += [de.mean(),de.std(),de.max() if len(de)>0 else 0,enhanced.mean(),enhanced.std()]
	gm = np.sqrt(ndimage.sobel(img,1)2+ndimage.sobel(img,0)2)
	dg = gm[dark_mask] if dark_mask.sum()>0 else np.array([0.0])
	features += [dg.mean(),dg.std(),dg.max() if len(dg)>0 else 0,gm.mean(),gm.std()]
	fi = np.abs(fftshift(fft2(img))); fd = np.abs(fftshift(fft2(img*dark_mask.astype(float))))
	features += [fi.mean(),fi.std(),fd.mean(),fd.std(),fd.sum()/(fi.sum()+1e-8)]
	from scipy.stats import skew,kurtosis
	features += [skew(dv),kurtosis(dv),np.percentile(img,10),np.percentile(img,25),np.median(img)-dv.mean()]
	lh = H//10
	for i in range(10): features.append(img[ilh:(i+1)lh,:].mean())
	return np.array(features[:35], dtype=np.float32)

	def extract_mvf(img):
	H,W = img.shape; features = []; bins = np.linspace(0,1,9); lc = []
	for i in range(8):
	mask = (img>=bins[i])&(img<bins[i+1]); lc.append(mask.sum()/(H*W))
	features += lc + [np.std(lc),max(lc)-min(lc)]
	tb = img.mean()
	for mult in [0.5,1.0,1.5,2.0,2.5]:
	er = (img>tb+multimg.std()).astype(float); features += [er.sum()/(HW),ndimage.label(er)[1]/max(1,H*W/1000)]
	for fog in [0.1,0.2,0.3,0.5,0.7]:
	vis = (img>fog).astype(float); features += [vis.mean(),ndimage.label(vis)[1]/max(1,H*W/500)]
	rh,rw = H//7,W//7
	for d in range(5):
	sm = ndimage.gaussian_filter(img,0.3+d*0.8)
	bm = [sm[irh:(i+1)rh,jrw:(j+1)rw].mean() for i in range(7) for j in range(7)]; features.append(np.std(bm))
	cy,cx = H//2,W//2; yy,xx = np.ogrid[:H,:W]; dist = np.sqrt((xx-cx)2+(yy-cy)2); mr = min(H,W)//2
	for rf in [0.1,0.25,0.5,0.75,0.95]:
	ring = (dist>=rfmr-3)&(dist<=rfmr+3); rv = img[ring] if ring.sum()>0 else np.array([0.0]); features.append(rv.std())
	return np.array(features[:40], dtype=np.float32)

	def extract_env(img):
	H,W = img.shape; features = []
	thresholds = {'black_hole':(0.0,0.10),'desert':(0.10,0.30),'clear_sea':(0.30,0.70),'red_sea':(0.70,0.90),'white_sea':(0.90,1.01)}
	masks = {n:(img>=lo)&(img<hi) for n,(lo,hi) in thresholds.items()}
	gm = np.sqrt(ndimage.sobel(img,0)2+ndimage.sobel(img,1)2)
	for n,mask in masks.items():
	cov = mask.sum()/(H*W); vals = img[mask] if mask.sum()>0 else np.array([0.0]); eg = gm[mask] if mask.sum()>0 else np.array([0.0])
	features += [cov,vals.mean(),vals.std(),eg.mean(),eg.std()]
	for n,mask in masks.items():
	labeled,nc = ndimage.label(mask.astype(int)); features += [nc/max(1,HW/1000),mask.sum()/max(nc,1)/(HW)]
	for i in range(5):
	row = int((i+0.5)*H/5); features.append(np.std(img[min(row,H-1),:]))
	return np.array(features[:40], dtype=np.float32)

	PHYSICS_TECHNIQUES = [
	('C11-DPM',extract_dpm,4,'🔥 تحليل الكثافة','Density Pattern Mismatch'),
	('C12-CKB',extract_ckb,8,'📋 قاعدة الكمارك','Customs Knowledge Base'),
	('C13-3DTW',extract_3dtw,25,'🌊 الموجة الحرارية','3D Thermal Wave'),
	('C14-CWS',extract_cws,50,'🌀 العاصفة الحلقية','Circular Wave Storm'),
	('C15-TCV',extract_tcv,42,'🌪️ الأعاصير الثلاثة','Triple Cyclone Vortex'),
	('C16-DWE',extract_dwe,35,'🔦 تعزيز المعتم','Dark Wave Enhancement'),
	('C17-MVF',extract_mvf,40,'🌋 طين-بركان-ضباب','Mud-Volcano-Fog'),
	('C18-ENV',extract_env,40,'🏜️ البيئات الخمس','Five Environments'),
	]
	PHYSICS_WEIGHTS = {'C11-DPM':0.10,'C12-CKB':0.10,'C13-3DTW':0.15,'C14-CWS':0.15,'C15-TCV':0.12,'C16-DWE':0.13,'C17-MVF':0.13,'C18-ENV':0.12}

	def physics_analyze(image):
	if image is None:
	empty = "<div style='text-align:center;padding:60px;color:#999;'>"
	empty += "<div style='font-size:48px;'>🔬</div>"
	empty += "<div>ارفع صورة X-Ray للتحليل الفيزيائي</div></div>"
	return empty, pd.DataFrame()
	start = time_module.time()
	img = physics_preprocess(image)
	if img is None:
	return "<div style='color:red;'>خطأ</div>", pd.DataFrame()
	# PRMI: Get RGB version for color analysis
	if isinstance(image, np.ndarray): _pil_img = Image.fromarray(image)
	else: _pil_img = image
	img_rgb_full = np.array(_pil_img.convert('RGB').resize((224,224)), dtype=np.float32)/255.0
	all_features = []; tech_results = {}
	H, W = img.shape
	quads = [img[:H//2,:W//2], img[:H//2,W//2:], img[H//2:,:W//2], img[H//2:,W//2:]]
	quad_means = [q.mean() for q in quads]
	quad_stds = [q.std() for q in quads]
	regional_var = np.std(quad_means)
	dark_ratio = (img < 0.2).sum() / (H*W)
	bright_ratio = (img > 0.8).sum() / (H*W)
	gm = np.sqrt(ndimage.sobel(img,0)2 + ndimage.sobel(img,1)2)
	edge_density = (gm > gm.mean() + 2gm.std()).sum() / (HW)
	strong_edge = (gm > gm.mean() + 3gm.std()).sum() / (HW)
	layers = [((img >= t) & (img < t+0.1)).sum()/(H*W) for t in np.linspace(0,1,10)]
	contrast = img.max() - img.min()
	bimodal = 0.0
	hist_vals = np.histogram(img, bins=20)[0].astype(float)
	hist_vals = hist_vals / hist_vals.sum()
	peaks = np.where((hist_vals[1:-1] > hist_vals[:-2]) & (hist_vals[1:-1] > hist_vals[2:]))[0]
	bimodal = len(peaks) / 10.0
	sharpness = strong_edge / (edge_density + 1e-8)
	metal_indicator = bright_ratio * 3.0 + sharpness * 2.0 + contrast * 1.5
	uniformity = 1.0 - np.std([q.std() for q in quads]) * 5.0
	uniformity = max(0, uniformity)
	cargo_normal = uniformity * 0.6 + (1 - regional_var * 5.0) * 0.4
	cargo_normal = float(np.clip(cargo_normal, 0, 1))
	# ML-based risk scoring (15 features)
	risk_features = [
	regional_var, dark_ratio, bright_ratio, edge_density, strong_edge,
	contrast, img.mean(), img.std(), np.std(layers),
	np.std([q.std() for q in quads]),
	np.max(quad_means) - np.min(quad_means),
	gm.mean(), gm.std(),
	len(np.where(np.diff(np.sign(np.diff(np.histogram(img,20)[0]))))[0]),
	np.percentile(img, 90) - np.percentile(img, 10),
	]
	if risk_model is not None:
	ml_risk_score = float(np.clip(risk_model.predict([risk_features])[0], 0, 1))
	else:
	ml_risk_score = float(np.clip(regional_var1.5 + dark_ratio0.8 + edge_density*1.0, 0, 1))
	img_suspicion = ml_risk_score
	for tc, func, dim, ar_name, en_name in PHYSICS_TECHNIQUES:
	try:
	feats = func(img)
	if len(feats) < dim: feats = np.pad(feats, (0, dim-len(feats)))
	feats = feats[:dim]; all_features.append(feats)
	nf = (feats - feats.min()) / (feats.max() - feats.min() + 1e-8)
	tech_var = nf.std()
	threat_boost = bright_ratio3.0 + strong_edge5.0 + (img.max()-img.min())*0.5
	threat_boost = float(np.clip(threat_boost, 0, 1))
	if ml_risk_score < 0.25:
	base = ml_risk_score * 0.9
	elif ml_risk_score < 0.45:
	base = ml_risk_score * 0.85 + threat_boost * 0.15
	else:
	base = max(ml_risk_score, threat_boost * 0.7)
	if tc == 'C11-DPM': score = float(np.clip(base0.85 + regional_var2.0 + tech_var*0.2, 0, 1))
	elif tc == 'C12-CKB': score = float(np.clip(base0.80 + tech_var0.3 + threat_boost*0.2, 0, 1))
	elif tc == 'C13-3DTW': score = float(np.clip(base0.75 + strong_edge4.0 + tech_var*0.2, 0, 1))
	elif tc == 'C14-CWS': score = float(np.clip(base0.80 + np.std(quad_stds)2.0 + threat_boost*0.15, 0, 1))
	elif tc == 'C15-TCV': score = float(np.clip(base0.80 + abs(nf[len(nf)//3:].mean()-nf[:len(nf)//3].mean())2.0, 0, 1))
	elif tc == 'C16-DWE': score = float(np.clip(base0.75 + dark_ratio2.5 + threat_boost*0.2, 0, 1))
	elif tc == 'C17-MVF': score = float(np.clip(base0.80 + np.std(layers)2.0 + threat_boost*0.15, 0, 1))
	elif tc == 'C18-ENV': score = float(np.clip(base0.80 + np.std(layers)1.5 + threat_boost*0.15, 0, 1))
	else: score = float(np.clip(base, 0, 1))
	tech_results[tc] = {'dim':dim,'ar':ar_name,'en':en_name,'score':score,'features':nf}
	except:
	all_features.append(np.zeros(dim))
	tech_results[tc] = {'dim':dim,'ar':ar_name,'en':en_name,'score':0,'features':np.zeros(dim)}
	fv = np.concatenate(all_features); fv = (fv-fv.min())/(fv.max()-fv.min()+1e-8)
	elapsed = time_module.time()-start
	final_score = sum(tech_results[k]['score']*PHYSICS_WEIGHTS[k] for k in tech_results)
	final_score = float(np.clip(final_score, 0, 1))
	if final_score >= 0.70: rl,rc,ri = 'حرج','#dc2626','🔴'
	elif final_score >= 0.50: rl,rc,ri = 'عالي','#ea580c','🟠'
	elif final_score >= 0.30: rl,rc,ri = 'متوسط','#ca8a04','🟡'
	else: rl,rc,ri = 'منخفض','#16a34a','🟢'
	fig = plt.figure(figsize=(14, 28)); fig.patch.set_facecolor('#0a1628')
	cl8 = ['#E53935','#FB8C00','#FDD835','#43A047','#1E88E5','#8E24AA','#F4511E','#00897B']
	ax1 = fig.add_subplot(9, 2, 1)
	names = [tc.split('-')[1] for tc in tech_results]; scores = [info['score'] for info in tech_results.values()]
	bars = ax1.barh(names, scores, color=cl8, edgecolor='white', linewidth=0.5)
	ax1.set_xlim(0, 1); ax1.set_facecolor('#0d1f3c'); ax1.tick_params(colors='white', labelsize=8)
	ax1.set_title('Technique Scores', color='#ffd740', fontsize=11, fontweight='bold')
	ax1.axvline(x=0.3, color='#FDD835', linestyle='--', alpha=0.5)
	ax1.axvline(x=0.5, color='#FB8C00', linestyle='--', alpha=0.5)
	ax1.axvline(x=0.7, color='#E53935', linestyle='--', alpha=0.5)
	for bar, s in zip(bars, scores): ax1.text(s+0.02, bar.get_y()+bar.get_height()/2, f'{s:.2f}', va='center', color='white', fontsize=8)
	ax2 = fig.add_subplot(9, 2, 2)
	ax2.imshow(img, cmap='hot', aspect='auto'); ax2.set_title('X-Ray Density Map', color='#ffd740', fontsize=11, fontweight='bold')
	ax2.set_facecolor('#0d1f3c'); ax2.tick_params(colors='white', labelsize=7)
	from scipy import ndimage as _ndi
	H, W = img.shape
	quads = [img[:H//2,:W//2], img[:H//2,W//2:], img[H//2:,:W//2], img[H//2:,W//2:]]
	gm = np.sqrt(_ndi.sobel(img,0)2 + _ndi.sobel(img,1)2)
	vis_titles_ar = ['تحليل كثافة الأرباع الأربعة','مطابقة 8 أنماط تهريب معروفة','تحليل الموجة الحرارية ثلاثية الأبعاد','مسح العاصفة الدائرية متعددة الحلقات','3 أعاصير حلزونية تمسح الصورة','تعزيز وإضاءة المناطق المعتمة','تحليل الطين والبركان والضباب','تصنيف 5 بيئات سطوعية مختلفة']
	tech_list = list(tech_results.items())
	for idx in range(8):
	tc, info = tech_list[idx]; nf = info['features']; sc = info['score']
	bc = '#E53935' if sc > 0.6 else '#FB8C00' if sc > 0.3 else '#43A047'
	ax_vis = fig.add_subplot(9, 2, 3 + idx*2)
	if idx == 0:
	quad_img = np.zeros_like(img)
	qm = [q.mean() for q in quads]
	quad_img[:H//2,:W//2] = qm[0]; quad_img[:H//2,W//2:] = qm[1]
	quad_img[H//2:,:W//2] = qm[2]; quad_img[H//2:,W//2:] = qm[3]
	ax_vis.imshow(quad_img, cmap='RdYlGn_r', aspect='auto')
	ax_vis.axhline(y=H//2, color='white', linewidth=2); ax_vis.axvline(x=W//2, color='white', linewidth=2)
	for qi, (qy,qx) in enumerate([(H//4,W//4),(H//4,3W//4),(3H//4,W//4),(3H//4,3W//4)]):
	ax_vis.text(qx, qy, f'Q{qi+1}:{qm[qi]:.2f}', ha='center', va='center', color='white', fontsize=8, fontweight='bold', bbox=dict(boxstyle='round', facecolor='black', alpha=0.6))
	elif idx == 1:
	susp = np.zeros_like(img); susp[img < 0.2] = 0.9; susp[img > 0.8] = 0.7; susp[(img > 0.3) & (img < 0.7)] = 0.2
	ax_vis.imshow(img * 0.5 + susp * 0.5, cmap='YlOrRd', aspect='auto')
	elif idx == 2:
	from scipy.ndimage import gaussian_filter
	thermal = gaussian_filter(img, sigma=3) - gaussian_filter(img, sigma=8)
	ax_vis.imshow(np.abs(thermal), cmap='inferno', aspect='auto')
	elif idx == 3:
	cy, cx = H//2, W//2; Y, X = np.mgrid[0:H, 0:W]
	dist = np.sqrt((X-cx)2 + (Y-cy)2)
	ax_vis.imshow(np.abs(np.sin(dist * 0.1) * img), cmap='plasma', aspect='auto')
	elif idx == 4:
	ax_vis.imshow(np.abs(img - np.fliplr(img)), cmap='coolwarm', aspect='auto')
	elif idx == 5:
	dark_e = np.zeros_like(img); dm = img < 0.3; dark_e[dm] = 1.0 - img[dm]; dark_e[~dm] = img[~dm] * 0.3
	ax_vis.imshow(dark_e, cmap='bone_r', aspect='auto')
	elif idx == 6:
	rgb = np.zeros((H, W, 3)); rgb[:,:,0] = np.clip((img < 0.3).astype(float) * 2, 0, 1)
	rgb[:,:,1] = np.clip(gm * 5, 0, 1); rgb[:,:,2] = np.clip((img > 0.7).astype(float) * 2, 0, 1)
	ax_vis.imshow(rgb, aspect='auto')
	elif idx == 7:
	env = np.digitize(img, [0.15, 0.35, 0.55, 0.75]) / 4.0
	ax_vis.imshow(env, cmap='viridis', aspect='auto')
	ax_vis.set_title(f'{tc} [{sc:.2f}]', color=bc, fontsize=9, fontweight='bold')
	ax_vis.set_xlabel(vis_titles_ar[idx], color='#aaa', fontsize=8)
	ax_vis.set_facecolor('#0d1f3c'); ax_vis.tick_params(colors='white', labelsize=6)
	ax_feat = fig.add_subplot(9, 2, 4 + idx*2)
	if idx == 0:
	ax_feat.bar(['Asym','StdD','Var','Dev'][:len(nf)], nf, color=cl8[0])
	elif idx == 1:
	ax_feat.bar(['Drug','Pill','Pwd','Liq','Mtl','Pls','Cig','Cash'][:len(nf)], nf, color=cl8[1])
	elif idx == 4:
	for k in range(min(3, len(nf)//14+1)):
	seg = nf[k14:(k+1)14]
	ax_feat.plot(seg, label=f'C{k+1}', color=cl8[4+k], linewidth=1.5)
	ax_feat.legend(fontsize=7, facecolor='#0d1f3c', labelcolor='white')
	else:
	ax_feat.plot(nf, color=cl8[idx], linewidth=1.5); ax_feat.fill_between(range(len(nf)), nf, alpha=0.3, color=cl8[idx])
	ax_feat.set_title(f'{tc} Features [{sc:.2f}]', color=bc, fontsize=9, fontweight='bold')
	ax_feat.set_facecolor('#0d1f3c'); ax_feat.tick_params(colors='white', labelsize=6)
	plt.tight_layout(pad=1.5)
	buf = io.BytesIO(); fig.savefig(buf, format='png', dpi=100, bbox_inches='tight', facecolor='#0a1628'); buf.seek(0)
	img_b64 = base64.b64encode(buf.read()).decode('utf-8'); plt.close(fig)
	# ═══ Generate 8 Animated GIFs ═══
	from PIL import Image as PILImage
	gif_b64_list = []
	H_g, W_g = img.shape
	gm_g = np.sqrt(ndimage.sobel(img,0)2 + ndimage.sobel(img,1)2)
	n_frames = 12
	for tidx in range(8):
	pil_frames = []
	for fi in range(n_frames):
	t = fi / (n_frames - 1)
	canvas = np.zeros((H_g, W_g, 3), dtype=np.uint8)
	if tidx == 0: # DPM - quadrant scan
	scan_y = int(t * H_g)
	vis = (img * 255).astype(np.uint8)
	canvas[:,:,0] = vis; canvas[:,:,1] = vis; canvas[:,:,2] = vis
	if scan_y > 0:
	above = img[:scan_y, :]
	qm = above.mean()
	heat = np.clip((np.abs(img[:scan_y,:] - qm) * 5 * 255), 0, 255).astype(np.uint8)
	canvas[:scan_y,:,0] = np.clip(vis[:scan_y,:].astype(int) + heat.astype(int), 0, 255).astype(np.uint8)
	canvas[:scan_y,:,1] = np.clip(vis[:scan_y,:].astype(int) - heat.astype(int)//2, 0, 255).astype(np.uint8)
	canvas[:scan_y,:,2] = vis[:scan_y,:]
	canvas[max(0,scan_y-2):min(H_g,scan_y+2),:,1] = 255
	elif tidx == 1: # CKB - pattern matching scan
	vis = (img * 255).astype(np.uint8)
	scan_x = int(t * W_g)
	canvas[:,:,0] = vis; canvas[:,:,1] = vis; canvas[:,:,2] = vis
	if scan_x > 0:
	region = img[:, :scan_x]
	dark_m = (region < 0.2)
	bright_m = (region > 0.8)
	red = np.zeros_like(region, dtype=np.uint8)
	red[dark_m] = 255
	orange = np.zeros_like(region, dtype=np.uint8)
	orange[bright_m] = 200
	canvas[:,:scan_x,0] = np.clip(vis[:,:scan_x].astype(int) + red.astype(int), 0, 255).astype(np.uint8)
	canvas[:,:scan_x,1] = np.clip(orange.astype(int), 0, 255).astype(np.uint8)
	canvas[:,:scan_x,2] = 0
	canvas[:,max(0,scan_x-2):min(W_g,scan_x+2),2] = 255
	elif tidx == 2: # 3DTW - thermal wave sweep
	from scipy.ndimage import gaussian_filter
	sigma = 2 + t * 8
	thermal = np.abs(gaussian_filter(img, sigma=max(1,sigma-3)) - gaussian_filter(img, sigma=sigma))
	thermal_n = np.clip(thermal / (thermal.max()+1e-8) * 255, 0, 255).astype(np.uint8)
	canvas[:,:,0] = thermal_n
	canvas[:,:,1] = np.clip(thermal_n // 2, 0, 255).astype(np.uint8)
	canvas[:,:,2] = 0
	elif tidx == 3: # CWS - expanding circular wave
	cy, cx = H_g//2, W_g//2
	Y, X = np.mgrid[0:H_g, 0:W_g]
	dist = np.sqrt((X-cx)2 + (Y-cy)2)
	max_r = np.sqrt(cx2 + cy2)
	radius = t * max_r
	wave = np.sin((dist - radius) * 0.15) * np.exp(-np.abs(dist - radius) / (max_r*0.15))
	wave_vis = np.clip((wave * 0.5 + 0.5) * img * 255, 0, 255).astype(np.uint8)
	ring_mask = (np.abs(dist - radius) < 5)
	canvas[:,:,0] = wave_vis
	canvas[:,:,1] = np.clip(wave_vis // 2, 0, 255).astype(np.uint8)
	canvas[:,:,2] = wave_vis
	canvas[ring_mask, 1] = 255
	elif tidx == 4: # TCV - 3 rotating cyclones
	angle = t * 2 * np.pi
	Y, X = np.mgrid[0:H_g, 0:W_g]
	combined = np.zeros((H_g, W_g))
	centers = [(H_g//4, W_g//4), (H_g//4, 3W_g//4), (3H_g//4, W_g//2)]
	for ci, (cy, cx) in enumerate(centers):
	dx = X - cx; dy = Y - cy
	r = np.sqrt(dx2 + dy2) + 1e-8
	theta = np.arctan2(dy, dx) + angle + ci*2.094
	spiral = np.sin(r0.05 + theta3) * np.exp(-r/(max(H_g,W_g)*0.3))
	combined += spiral
	combined = np.clip((combined * 0.3 + 0.5) * img, 0, 1)
	canvas[:,:,0] = np.clip(combined * 200, 0, 255).astype(np.uint8)
	canvas[:,:,1] = np.clip(combined * 100, 0, 255).astype(np.uint8)
	canvas[:,:,2] = np.clip(combined * 255, 0, 255).astype(np.uint8)
	elif tidx == 5: # DWE - light sweeping dark regions
	vis = (img * 255).astype(np.uint8)
	sweep_x = int(t * W_g)
	dark_mask = img < 0.3
	canvas[:,:,0] = vis; canvas[:,:,1] = vis; canvas[:,:,2] = vis
	if sweep_x > 0:
	reveal = np.zeros((H_g, sweep_x), dtype=np.uint8)
	dm_region = dark_mask[:, :sweep_x]
	enhanced = np.clip((1.0 - img[:,:sweep_x]) * 255, 0, 255).astype(np.uint8)
	canvas[:,:sweep_x,0] = np.where(dm_region, enhanced, vis[:,:sweep_x])
	canvas[:,:sweep_x,1] = np.where(dm_region, enhanced, vis[:,:sweep_x])
	canvas[:,:sweep_x,2] = np.where(dm_region, np.clip(enhanced//2, 0, 255), vis[:,:sweep_x])
	canvas[:,max(0,sweep_x-3):min(W_g,sweep_x+3),0] = 255
	canvas[:,max(0,sweep_x-3):min(W_g,sweep_x+3),1] = 255
	canvas[:,max(0,sweep_x-3):min(W_g,sweep_x+3),2] = 0
	elif tidx == 6: # MVF - mud(red) volcano(green) fog(blue) layers
	vis = (img * 255).astype(np.uint8)
	layer = int(t * 3)
	canvas[:,:,0] = vis; canvas[:,:,1] = vis; canvas[:,:,2] = vis
	if layer >= 0:
	mud = np.clip((img < 0.3).astype(float) * (1-img) * 255 * min(t*3, 1), 0, 255).astype(np.uint8)
	canvas[:,:,0] = np.clip(vis.astype(int) + mud.astype(int), 0, 255).astype(np.uint8)
	if layer >= 1:
	volc = np.clip(gm_g * 5 * 255 * min((t-0.33)*3, 1), 0, 255).astype(np.uint8)
	canvas[:,:,1] = np.clip(vis.astype(int) + volc.astype(int), 0, 255).astype(np.uint8)
	if layer >= 2:
	fog = np.clip((img > 0.7).astype(float) * img * 255 * min((t-0.66)*3, 1), 0, 255).astype(np.uint8)
	canvas[:,:,2] = np.clip(vis.astype(int) + fog.astype(int), 0, 255).astype(np.uint8)
	elif tidx == 7: # ENV - 5 environments appearing one by one
	env_step = int(t * 5)
	canvas_f = np.zeros((H_g, W_g, 3))
	colors = [(0.2,0,0.4), (0,0.2,0.5), (0,0.5,0.3), (0.5,0.5,0), (0.8,0.3,0)]
	thresholds = [0.15, 0.35, 0.55, 0.75, 1.01]
	prev_t = 0
	for ei in range(min(env_step+1, 5)):
	mask = (img >= prev_t) & (img < thresholds[ei])
	for ch in range(3):
	canvas_f[:,:,ch] = np.where(mask, colors[ei][ch] + img*0.5, canvas_f[:,:,ch])
	prev_t = thresholds[ei]
	canvas = np.clip(canvas_f * 255, 0, 255).astype(np.uint8)
	pil_frames.append(PILImage.fromarray(canvas))
	gif_buf = io.BytesIO()
	pil_frames[0].save(gif_buf, format='GIF', save_all=True, append_images=pil_frames[1:], duration=250, loop=0)
	gif_buf.seek(0)
	gif_b64_list.append(base64.b64encode(gif_buf.read()).decode('utf-8'))
	gif_names = ['C11-DPM: مسح الكثافة','C12-CKB: كشف أنماط التهريب','C13-3DTW: الموجة الحرارية','C14-CWS: العاصفة الدائرية','C15-TCV: الأعاصير الثلاثة','C16-DWE: إضاءة الظلام','C17-MVF: طين+بركان+ضباب','C18-ENV: البيئات الخمسة']
	h = "<div style='font-family:Tajawal,sans-serif;direction:rtl;'>"
	h += f"<div style='background:linear-gradient(135deg,#0D47A1,#1565C0);padding:12px;border-radius:12px;text-align:center;color:white;margin-bottom:10px;'>"
	h += f"<div style='font-size:13px;'>🔬 التحليل الفيزيائي — 8 تقنيات — {len(fv)} بُعد</div></div>"
	h += f"<div style='background:{rc};padding:16px;border-radius:12px;text-align:center;color:white;margin-bottom:10px;'>"
	h += f"<div style='font-size:24px;font-weight:bold;'>{ri} خطورة الإخفاء: {rl} ({final_score:.1%})</div></div>"
	h += "<div style='display:grid;grid-template-columns:repeat(4,1fr);gap:6px;margin-bottom:10px;'>"
	for val,lbl,bg in [(f'{len(fv)}D','الأبعاد','#E3F2FD'),('8','تقنيات','#F3E5F5'),(f'{elapsed:.1f}s','الوقت','#E8F5E9'),(f'{fv.mean():.3f}','المتوسط','#FFF3E0')]:
	h += f"<div style='background:{bg};padding:10px;border-radius:10px;text-align:center;'><div style='font-size:22px;font-weight:bold;'>{val}</div><div style='color:#666;font-size:11px;'>{lbl}</div></div>"
	h += "</div>"
	h += f"<div style='margin:10px 0;border-radius:12px;overflow:hidden;border:2px solid #1565C0;'><img src='data:image/png;base64,{img_b64}' style='width:100%;'/></div>"
	h += "<div style='background:linear-gradient(135deg,#1a237e,#0d47a1);padding:12px;border-radius:12px;text-align:center;color:white;margin:10px 0;font-size:16px;font-weight:bold;'>🌊 التحليل الديناميكي — 8 تقنيات متحركة</div>"
	h += "<div style='display:grid;grid-template-columns:repeat(2,1fr);gap:8px;margin:10px 0;'>"
	for gi in range(8):
	gsc = list(tech_results.values())[gi]['score']
	gbc = '#dc2626' if gsc > 0.6 else '#ea580c' if gsc > 0.3 else '#16a34a'
	h += f"<div style='background:#0d1f3c;border-radius:10px;overflow:hidden;border:2px solid {gbc};'>"
	h += f"<div style='background:{gbc};padding:6px;text-align:center;color:white;font-size:13px;font-weight:bold;'>{gif_names[gi]} [{gsc:.0%}]</div>"
	h += f"<img src='data:image/gif;base64,{gif_b64_list[gi]}' style='width:100%;'/></div>"
	h += "</div>"
	old_chart_img = "SKIP"
	h += f"<div style='margin:10px 0;border-radius:12px;overflow:hidden;border:2px solid #1565C0;'><img src='data:image/png;base64,{img_b64}' style='width:100%;display:block;'/></div>"
	h += "<table style='width:100%;border-collapse:collapse;font-size:12px;'>"
	h += "<tr style='background:#1565C0;color:white;'><th style='padding:8px;text-align:right;'>التقنية</th><th style='padding:8px;text-align:center;'>EN</th><th style='padding:8px;text-align:center;'>الأبعاد</th><th style='padding:8px;text-align:center;'>الدرجة</th><th style='padding:8px;text-align:center;'>مؤشر</th></tr>"
	tcl = ['#1565C0','#2E7D32','#6A1B9A','#E65100','#00838F','#D81B60','#F57C00','#00695C']
	for idx,(tc,info) in enumerate(tech_results.items()):
	bg = '#f8f9fa' if idx%2==0 else '#fff'; sc = info['score']; bw = int(sc*100)
	bc2 = '#E53935' if sc>0.6 else '#FB8C00' if sc>0.3 else '#43A047'
	h += f"<tr style='background:{bg};border-bottom:1px solid #e0e0e0;'><td style='padding:6px 8px;font-weight:bold;color:{tcl[idx]};'>({tc}) {info['ar']}</td><td style='padding:6px;text-align:center;font-size:10px;color:#666;'>{info['en']}</td><td style='padding:6px;text-align:center;font-weight:bold;'>{info['dim']}D</td><td style='padding:6px;text-align:center;'>{sc:.3f}</td><td style='padding:6px;'><div style='background:#eee;border-radius:10px;overflow:hidden;height:16px;'><div style='background:{bc2};height:100%;width:{bw}%;border-radius:10px;'></div></div></td></tr>"
	h += "</table>"
	h += "<div style='background:#1a1a2e;padding:15px;border-radius:12px;margin-top:10px;border:1px solid #333;direction:rtl;'>"
	h += "<div style='text-align:center;color:#ffd740;font-weight:bold;font-size:14px;margin-bottom:10px;'>📊 دليل الألوان ومستويات الخطورة</div>"
	h += "<div style='display:grid;grid-template-columns:repeat(4,1fr);gap:8px;margin-bottom:12px;'>"
	h += "<div style='background:#16a34a;color:white;padding:8px;border-radius:8px;text-align:center;'><div style='font-size:16px;font-weight:bold;'>🟢 منخفض</div><div style='font-size:11px;'>0% - 30%</div><div style='font-size:10px;'>بضاعة طبيعية</div></div>"
	h += "<div style='background:#ca8a04;color:white;padding:8px;border-radius:8px;text-align:center;'><div style='font-size:16px;font-weight:bold;'>🟡 متوسط</div><div style='font-size:11px;'>30% - 50%</div><div style='font-size:10px;'>يحتاج مراجعة</div></div>"
	h += "<div style='background:#ea580c;color:white;padding:8px;border-radius:8px;text-align:center;'><div style='font-size:16px;font-weight:bold;'>🟠 عالي</div><div style='font-size:11px;'>50% - 70%</div><div style='font-size:10px;'>فحص يدوي</div></div>"
	h += "<div style='background:#dc2626;color:white;padding:8px;border-radius:8px;text-align:center;'><div style='font-size:16px;font-weight:bold;'>🔴 حرج</div><div style='font-size:11px;'>70% - 100%</div><div style='font-size:10px;'>إيقاف وتفتيش</div></div>"
	h += "</div>"
	h += "<div style='color:#ccc;font-size:11px;line-height:1.8;'>"
	h += "<div style='display:grid;grid-template-columns:1fr 1fr;gap:6px;'>"
	h += "<div>🔥 <b style='color:#E53935;'>DPM</b>: يقيس فروقات الكثافة بين أرباع الصورة — كثافة غير متجانسة = مشبوه</div>"
	h += "<div>📋 <b style='color:#FB8C00;'>CKB</b>: يطابق 8 أنماط تهريب عراقية (مخدرات/حبوب/مسحوق/سائل/معدن/بلاستيك/سجائر/نقود)</div>"
	h += "<div>🌊 <b style='color:#FDD835;'>3DTW</b>: يحوّل الصورة لسطح ثلاثي الأبعاد ويكشف الموجات الحرارية الشاذة</div>"
	h += "<div>🌀 <b style='color:#43A047;'>CWS</b>: يمسح 8 حلقات دائرية + 4 عواصف اتجاهية لكشف الأنماط المخفية</div>"
	h += "<div>🌪️ <b style='color:#1E88E5;'>TCV</b>: 3 أعاصير حلزونية تأخذ عينات من يسار/وسط/يمين الصورة</div>"
	h += "<div>🔦 <b style='color:#8E24AA;'>DWE</b>: يعزز المناطق المعتمة حيث يختبئ التهريب عادةً</div>"
	h += "<div>🌋 <b style='color:#F4511E;'>MVF</b>: يحلل طبقات الكثافة (طين) + مستويات الانفجار (بركان) + الرؤية (ضباب)</div>"
	h += "<div>🏜️ <b style='color:#00897B;'>ENV</b>: يقسم الصورة إلى 5 بيئات سطوعية ويحلل كل واحدة</div>"
	h += "</div></div>"
	h += "<div style='text-align:center;margin-top:8px;color:#888;font-size:10px;'>المعادلة: C = 0.10×DPM + 0.10×CKB + 0.15×3DTW + 0.15×CWS + 0.12×TCV + 0.13×DWE + 0.13×MVF + 0.12×ENV</div>"
	h += "</div>"
	h += f"<div style='background:#E8F5E9;padding:10px;border-radius:10px;margin-top:8px;border:1px solid #A5D6A7;text-align:center;font-size:12px;'><b>📊</b> {len(fv)}D \| Mean: {fv.mean():.4f} \| Std: {fv.std():.4f} \| Non-zero: {(fv!=0).sum()}/{len(fv)}</div></div>"

	# ═══ C19-VTA: Visual Threat Alert System ═══
	from PIL import Image as PILImage, ImageDraw, ImageFont
	from scipy import ndimage as vta_ndi
	H_v, W_v = img.shape
	grid_rows, grid_cols = 8, 12
	cell_h, cell_w = H_v // grid_rows, W_v // grid_cols
	densities_grid = np.zeros((grid_rows, grid_cols))
	variance_grid = np.zeros((grid_rows, grid_cols))
	edge_grid = np.zeros((grid_rows, grid_cols))
	gm_vta = np.sqrt(vta_ndi.sobel(img,0)2 + vta_ndi.sobel(img,1)2)
	for r in range(grid_rows):
	for c in range(grid_cols):
	cell = img[rcell_h:(r+1)cell_h, ccell_w:(c+1)cell_w]
	cell_edge = gm_vta[rcell_h:(r+1)cell_h, ccell_w:(c+1)cell_w]
	densities_grid[r, c] = cell.mean()
	variance_grid[r, c] = cell.std()
	edge_grid[r, c] = cell_edge.mean()
	g_mean = densities_grid.mean()
	g_std = densities_grid.std()
	TH_CAUTION = 0.1644
	TH_WARNING = 0.1762
	TH_CRITICAL = 0.2229
	TH_HIGH_DENSITY = 0.4606
	alerts = []
	for r in range(grid_rows):
	for c in range(grid_cols):
	d = densities_grid[r, c]
	v = variance_grid[r, c]
	e = edge_grid[r, c]
	neighbors = []
	for dr, dc in [(-1,0),(1,0),(0,-1),(0,1),(-1,-1),(-1,1),(1,-1),(1,1)]:
	nr, nc = r+dr, c+dc
	if 0 <= nr < grid_rows and 0 <= nc < grid_cols:
	neighbors.append(densities_grid[nr, nc])
	n_mean = np.mean(neighbors) if neighbors else g_mean
	n_diff = abs(d - n_mean)
	is_border = r == 0 or r == grid_rows-1 or c == 0 or c == grid_cols-1
	level = 'safe'
	if is_border:
	level = 'safe'
	elif n_diff > TH_CRITICAL and d > TH_HIGH_DENSITY:
	level = 'critical'
	elif n_diff > TH_CRITICAL:
	level = 'warning'
	elif n_diff > TH_WARNING and d > TH_HIGH_DENSITY:
	level = 'warning'
	elif n_diff > TH_CAUTION and d > TH_HIGH_DENSITY and e > 0.3:
	level = 'caution'
	if level != 'safe':
	alerts.append({'r':r,'c':c,'density':d,'n_diff':n_diff,'var':v,'level':level,
	'x1':ccell_w,'y1':rcell_h,'x2':(c+1)cell_w,'y2':(r+1)cell_h})
	img_rgb = np.stack([np.clip(img255,0,255).astype(np.uint8)]3, axis=-1)
	vta_pil = PILImage.fromarray(img_rgb)
	draw = ImageDraw.Draw(vta_pil)
	colors_map = {'critical':'#FF0000','warning':'#FFA500','caution':'#FFFF00'}
	widths_map = {'critical':4,'warning':3,'caution':2}
	icons_map = {'critical':'🔴','warning':'🟡','caution':'🟢'}
	for alert in alerts:
	color = colors_map[alert['level']]
	width = widths_map[alert['level']]
	x1,y1,x2,y2 = alert['x1'],alert['y1'],alert['x2'],alert['y2']
	draw.rectangle([x1,y1,x2,y2], outline=color, width=width)
	if alert['level'] == 'critical':
	for offset in range(-2,3):
	draw.rectangle([x1+offset,y1+offset,x2-offset,y2-offset], outline=color, width=1)
	label = f"{alert['density']:.0%}"
	draw.text((x1+4, y1+4), label, fill=color)
	critical_count = sum(1 for a in alerts if a['level']=='critical')
	warning_count = sum(1 for a in alerts if a['level']=='warning')
	caution_count = sum(1 for a in alerts if a['level']=='caution')
	total_threat = (critical_count * 3 + warning_count * 2 + caution_count * 1)
	max_threat = grid_rows * grid_cols * 3
	threat_pct = min(total_threat / max(max_threat * 0.15, 1), 1.0)
	if critical_count > 0:
	badge_color = '#FF0000'
	badge_text = f'⛔ {critical_count} مناطق حرجة'
	elif warning_count > 0:
	badge_color = '#FFA500'
	badge_text = f'⚠️ {warning_count} مناطق مشبوهة'
	else:
	badge_color = '#16a34a'
	badge_text = '✅ لا توجد تهديدات'
	badge_h = min(80, H_v // 8)
	draw.rectangle([0, 0, W_v, badge_h], fill='#000000')
	draw.text((10, 5), badge_text, fill=badge_color)
	draw.text((10, badge_h//2), f'Threat: {threat_pct:.0%} \| Critical:{critical_count} Warning:{warning_count} Caution:{caution_count}', fill='white')
	vta_frames = []
	for fi in range(15):
	frame = vta_pil.copy()
	fdraw = ImageDraw.Draw(frame)
	pulse = abs(np.sin(fi * np.pi / 7))
	for alert in alerts:
	if alert['level'] == 'critical':
	color_r = int(255 * pulse)
	color_hex = f'#{color_r:02x}0000'
	x1,y1,x2,y2 = alert['x1'],alert['y1'],alert['x2'],alert['y2']
	expand = int(3 * pulse)
	fdraw.rectangle([x1-expand,y1-expand,x2+expand,y2+expand], outline=color_hex, width=3)
	cx, cy = (x1+x2)//2, (y1+y2)//2
	radius = int(min(cell_w, cell_h) * 0.4 * pulse)
	fdraw.ellipse([cx-radius, cy-radius, cx+radius, cy+radius], outline='#FF0000', width=2)
	elif alert['level'] == 'warning':
	if fi % 3 == 0:
	x1,y1,x2,y2 = alert['x1'],alert['y1'],alert['x2'],alert['y2']
	fdraw.rectangle([x1,y1,x2,y2], outline='#FFA500', width=3)
	vta_frames.append(frame)
	vta_gif_buf = io.BytesIO()
	vta_frames[0].save(vta_gif_buf, format='GIF', save_all=True, append_images=vta_frames[1:], duration=300, loop=0)
	vta_gif_buf.seek(0)
	vta_gif_b64 = base64.b64encode(vta_gif_buf.read()).decode('utf-8')
	vta_static_buf = io.BytesIO()
	vta_pil.save(vta_static_buf, format='PNG')
	vta_static_buf.seek(0)
	vta_static_b64 = base64.b64encode(vta_static_buf.read()).decode('utf-8')
	h += "<div style='background:linear-gradient(135deg,#b71c1c,#d32f2f);padding:12px;border-radius:12px;text-align:center;color:white;margin:10px 0;'>"
	h += "<div style='font-size:18px;font-weight:bold;'>🚨 C19-VTA: نظام الإنذار البصري المروري</div>"
	h += "<div style='font-size:12px;margin-top:4px;'>Visual Threat Alert — Traffic Light Annotation System</div></div>"
	h += "<div style='display:grid;grid-template-columns:repeat(4,1fr);gap:6px;margin:8px 0;'>"
	for val,lbl,bg in [(f'{critical_count}','حرج 🔴','#FFCDD2'),(f'{warning_count}','تحذير 🟡','#FFF9C4'),(f'{caution_count}','تنبيه 🟢','#E8F5E9'),(f'{threat_pct:.0%}','مستوى التهديد','#E3F2FD')]:
	h += f"<div style='background:{bg};padding:8px;border-radius:8px;text-align:center;'><div style='font-size:20px;font-weight:bold;'>{val}</div><div style='font-size:11px;'>{lbl}</div></div>"
	h += "</div>"
	h += f"<div style='margin:8px 0;border-radius:12px;overflow:hidden;border:3px solid {badge_color};'>"
	h += f"<div style='background:#000;padding:6px;text-align:center;color:{badge_color};font-weight:bold;font-size:14px;'>{badge_text}</div>"
	h += f"<img src='data:image/gif;base64,{vta_gif_b64}' style='width:100%;'/></div>"
	h += "<div style='margin:8px 0;border-radius:12px;overflow:hidden;border:2px solid #1565C0;'>"
	h += "<div style='background:#0d47a1;padding:6px;text-align:center;color:white;font-weight:bold;font-size:13px;'>📸 خريطة التهديد الثابتة</div>"
	h += f"<img src='data:image/png;base64,{vta_static_b64}' style='width:100%;'/></div>"
	if alerts:
	h += "<div style='background:#1a1a2e;padding:10px;border-radius:10px;margin:8px 0;'>"
	h += "<div style='color:#ffd740;font-weight:bold;margin-bottom:6px;'>📋 تفاصيل المناطق المشبوهة:</div>"
	for i, alert in enumerate(sorted(alerts, key=lambda x: -{'critical':3,'warning':2,'caution':1}[x['level']])):
	icon = icons_map[alert['level']]
	h += f"<div style='color:white;font-size:12px;padding:3px 0;'>{icon} المنطقة [{alert['r']},{alert['c']}] — الكثافة: {alert['density']:.1%} — فرق الجيران: {alert['n_diff']:.3f} — التباين: {alert['var']:.3f}</div>"
	h += "</div>"
	h += "<div style='background:#E3F2FD;padding:8px;border-radius:8px;margin:8px 0;font-size:11px;'>"
	h += "<b>🚦 دليل الإشارات المرورية:</b><br>"
	h += "🔴 <b>حرج</b>: كثافة أدوية/مخدرات + تباين مفاجئ مع الجيران (مربع أحمر نابض)<br>"
	h += "🟡 <b>تحذير</b>: كثافة مشبوهة متجانسة أو انحراف كبير عن المعدل (مربع برتقالي)<br>"
	h += "🟢 <b>تنبيه</b>: تباين خفيف يستحق المراجعة (مربع أصفر)<br>"
	h += "✅ <b>آمن</b>: كثافة طبيعية متوافقة مع البضاعة المصرّحة</div>"
	# ═══ PRMI v7: Direction-Aware Material Analysis ═══
	try:
	import pickle as prmi_pickle
	from scipy.stats import skew as sp_skew
	prmi_path = os.path.join(os.path.dirname(__file__), 'prmi_model.pkl')
	if not os.path.exists(prmi_path):
	from huggingface_hub import hf_hub_download
	prmi_path = hf_hub_download(repo_id='DrAbbas/SONAR-AI', filename='prmi_model.pkl')
	with open(prmi_path, 'rb') as pf:
	prmi_data = prmi_pickle.load(pf)
	prmi_sigs = prmi_data.get('signatures', {})
	gm_prmi = np.sqrt(ndimage.sobel(img,0)2 + ndimage.sobel(img,1)2)
	prmi_gr, prmi_gc = 8, 12
	prmi_ch, prmi_cw = H_v // prmi_gr, W_v // prmi_gc
	# Extract 18D features for all cells
	all_cells = []
	prmi_results = []
	main_mat = 'unknown'
	from collections import Counter as PrmiCounter
	for r in range(1, prmi_gr-1):
	for c in range(1, prmi_gc-1):
	pg = img[rprmi_ch:(r+1)prmi_ch, cprmi_cw:(c+1)prmi_cw]
	pe = gm_prmi[rprmi_ch:(r+1)prmi_ch, cprmi_cw:(c+1)prmi_cw]
	d = pg.mean()
	if 0.05 < d < 0.95:
	t_v = pg.std()
	e_v = pe.mean()
	flat = pg.flatten()
	try:
	sk = float(sp_skew(flat))
	except:
	sk = 0.0
	hist_p, _ = np.histogram(pg, bins=16, range=(0,1))
	hist_p = hist_p / hist_p.sum() + 1e-10
	ent = float(-np.sum(hist_p * np.log2(hist_p)))
	nb = []
	for dr, dc in [(-1,0),(1,0),(0,-1),(0,1)]:
	nr, nc = r+dr, c+dc
	if 0<=nr<prmi_gr and 0<=nc<prmi_gc:
	nb.append(img[nrprmi_ch:(nr+1)prmi_ch, ncprmi_cw:(nc+1)prmi_cw].mean())
	ndiff = abs(d - np.mean(nb)) if nb else 0
	all_cells.append({'r':r,'c':c,'d':d,'t':t_v,'e':e_v,'skew':sk,'entropy':ent,'ndiff':ndiff})
	# Calculate image majority
	if all_cells:
	# Filter out background cells (Blue% > 40%) before computing stats
	fg_cells = [c for c in all_cells if c.get('bp', 0.33) <= 0.40]
	if not fg_cells:
	fg_cells = all_cells # fallback if all cells are background
	img_d_med = np.median([c['d'] for c in fg_cells])
	img_d_std = np.std([c['d'] for c in fg_cells]) + 0.005
	img_sk_med = np.median([c['skew'] for c in fg_cells])
	img_sk_std = np.std([c['skew'] for c in fg_cells]) + 0.01
	img_ent_med = np.median([c['entropy'] for c in fg_cells])
	img_ent_std = np.std([c['entropy'] for c in fg_cells]) + 0.01
	# Direction-aware + Color anomaly detection
	# Check if image is colored (R/G != 1.0)
	rgb_r_prmi = np.array(Image.fromarray((img_rgb_full*255).astype(np.uint8)).resize((W_v, H_v)), dtype=np.float32)/255.0
	rg_vals = []
	bp_vals = []
	for cell in all_cells:
	pr = rgb_r_prmi[cell['r']prmi_ch:(cell['r']+1)prmi_ch, cell['c']prmi_cw:(cell['c']+1)prmi_cw]
	R_p=pr[:,:,0].mean(); G_p=pr[:,:,1].mean(); B_p=pr[:,:,2].mean()
	total_p=R_p+G_p+B_p+1e-8
	cell['rg'] = R_p/(G_p+1e-8)
	cell['bp'] = B_p/total_p
	rg_vals.append(cell['rg'])
	bp_vals.append(cell['bp'])
	# Use only foreground cells for stats
	fg_rg = [c['rg'] for c in all_cells if c.get('bp', 0.33) <= 0.40]
	fg_bp = [c['bp'] for c in all_cells if c.get('bp', 0.33) <= 0.40]
	if not fg_rg:
	fg_rg = rg_vals
	fg_bp = bp_vals
	rg_med = np.median(fg_rg); rg_std_v = np.std(fg_rg) + 0.01
	bp_med = np.median(fg_bp); bp_std_v = np.std(fg_bp) + 0.01
	is_colored = abs(rg_med - 1.0) > 0.05
	# Find main material
	min_dist = 999
	e_med_v = np.median([c['e'] for c in fg_cells])
	t_med_v = np.median([c['t'] for c in fg_cells])
	ent_med_v = np.median([c['entropy'] for c in fg_cells])
	for cls, sig in prmi_sigs.items():
	if is_colored:
	dist = (abs(img_d_med - sig.get('d',0)) / (sig.get('d_std',0.1)+0.01) * 0.20 +
	abs(rg_med - sig.get('rg',1.0)) / (sig.get('rg_std',0.1)+0.01) * 0.30 +
	abs(bp_med - sig.get('bp',0.2)) / (sig.get('bp_std',0.05)+0.01) * 0.25 +
	abs(e_med_v - sig.get('e',0.3)) / (sig.get('e_std',0.1)+0.01) * 0.15 +
	abs(t_med_v - sig.get('t',0.1)) / (sig.get('t_std',0.05)+0.01) * 0.10)
	else:
	dist = (abs(img_d_med - sig.get('d',0)) / (sig.get('d_std',0.1)+0.01) * 0.35 +
	abs(e_med_v - sig.get('e',0.3)) / (sig.get('e_std',0.1)+0.01) * 0.25 +
	abs(t_med_v - sig.get('t',0.1)) / (sig.get('t_std',0.05)+0.01) * 0.20 +
	abs(ent_med_v - sig.get('ent',2.0)) / (sig.get('ent_std',0.5)+0.01) * 0.20)
	if dist < min_dist: min_dist = dist; main_mat = cls
	for cell in all_cells:
	# v8: Skip background cells
	if cell.get('bp', 0.33) > 0.40:
	continue
	z_d_low = (img_d_med - cell['d']) / img_d_std
	z_sk_high = (cell['skew'] - img_sk_med) / img_sk_std
	z_ent_low = (img_ent_med - cell['entropy']) / img_ent_std
	score_dir = z_d_low * 0.4 + z_sk_high * 0.4 + z_ent_low * 0.2
	z_rg = abs(cell.get('rg',1.0) - rg_med) / rg_std_v
	z_bp = abs(cell.get('bp',0.33) - bp_med) / bp_std_v
	score_col = z_rg * 0.5 + z_bp * 0.5
	score = score_dir * 0.3 + score_col * 0.7 if is_colored else score_dir
	if score > 3.0 or (is_colored and score_col > 4.0):
	# Find closest class for labeling
	cell_mat = 'unknown'
	cell_dist = 999
	for cls, sig in prmi_sigs.items():
	if is_colored:
	dist = (abs(cell['d'] - sig.get('d',0)) / (sig.get('d_std',0.1)+0.01) * 0.20 +
	abs(cell.get('rg',1.0) - sig.get('rg',1.0)) / (sig.get('rg_std',0.1)+0.01) * 0.30 +
	abs(cell.get('bp',0.33) - sig.get('bp',0.2)) / (sig.get('bp_std',0.05)+0.01) * 0.25 +
	abs(cell.get('e',0.3) - sig.get('e',0.3)) / (sig.get('e_std',0.1)+0.01) * 0.15 +
	abs(cell.get('t',0.1) - sig.get('t',0.1)) / (sig.get('t_std',0.05)+0.01) * 0.10)
	else:
	dist = (abs(cell['d'] - sig.get('d',0)) / (sig.get('d_std',0.1)+0.01) * 0.35 +
	abs(cell.get('e',0.3) - sig.get('e',0.3)) / (sig.get('e_std',0.1)+0.01) * 0.25 +
	abs(cell.get('t',0.1) - sig.get('t',0.1)) / (sig.get('t_std',0.05)+0.01) * 0.20 +
	abs(cell.get('entropy',2.0) - sig.get('ent',2.0)) / (sig.get('ent_std',0.5)+0.01) * 0.20)
	if dist < cell_dist: cell_dist = dist; cell_mat = cls
	level = 'critical' if score > 5.0 else 'warning' if score > 4.0 else 'caution'
	prmi_results.append({'r':cell['r'],'c':cell['c'],'pred':cell_mat,
	'score':score,'level':level,'density':cell['d'],
	'skew':cell['skew'],'entropy':cell['entropy']})
	# PRMI v8.3: NO boxes, NO map
	anomaly_cnt = len(prmi_results)
	mat_names_ar = {'banana':'موز','milk':'حليب','medical':'أدوية','weapons':'أسلحة','electronics':'إلكترونيات','clothes':'ملابس','chemicals':'كيماويات','steel':'فولاذ','glass':'زجاج','plastic':'بلاستيك'}
	h += "<div style='background:linear-gradient(135deg,#1a237e,#4A148C);padding:16px;border-radius:14px;margin:8px 0;'>"
	h += "<div style='text-align:center;color:white;font-size:14px;margin-bottom:10px;'>🧬 PRMI v8 (43 مادة)</div>"
	if anomaly_cnt > 0:
	anom_mats = list(set(pr['pred'] for pr in prmi_results))[:3]
	anom_ar = [mat_names_ar.get(m, m) for m in anom_mats]
	h += "<div style='background:rgba(244,67,54,0.2);border-radius:12px;padding:16px;text-align:center;'>"
	h += "<div style='color:#EF9A9A;font-size:18px;'>⚠️ مواد مخفية: " + ' + '.join(anom_ar) + "</div>"
	h += "<div style='color:#EF9A9A;font-size:12px;'>" + str(anomaly_cnt) + " منطقة</div>"
	h += "</div>"
	else:
	h += "<div style='background:rgba(76,175,80,0.2);border-radius:12px;padding:16px;text-align:center;'>"
	h += "<div style='color:#A5D6A7;font-size:20px;font-weight:bold;'>✅ الحاوية متجانسة — لا مواد مخفية</div>"
	h += "</div>"
	h += "</div>"
	except Exception as prmi_err:
	h += f"<div style='background:#FF0000;padding:10px;border-radius:8px;margin:8px 0;color:white;'>🔴 PRMI Error: {str(prmi_err)}</div>"

	rows = [{'#':tc,'التقنية':info['ar'],'EN':info['en'],'الأبعاد':f"{info['dim']}D",'الدرجة':f"{info['score']:.4f}",'الوزن':f"{PHYSICS_WEIGHTS[tc]:.2f}"} for tc,info in tech_results.items()]
	return h, pd.DataFrame(rows)

	import requests as http_requests

	def fetch_db_stats():
	"""جلب إحصائيات من السيرفر"""
	try:
	r = http_requests.get(f"{API_URL}/stats", timeout=5)
	if r.status_code == 200:
	return r.json()
	except:
	pass
	return None

	def fetch_db_anomalies():
	"""جلب آخر المخالفات"""
	try:
	r = http_requests.get(f"{API_URL}/anomalies?limit=20", timeout=5)
	if r.status_code == 200:
	data = r.json()
	if data:
	return pd.DataFrame(data)
	except:
	pass
	return None

	def refresh_dashboard():
	"""تحديث لوحة التحكم من قاعدة البيانات"""
	stats = fetch_db_stats()
	if stats:
	return generate_dashboard_html(stats), get_anomalies_from_db()
	return generate_dashboard_html(None), get_anomalies()

	def get_anomalies_from_db():
	"""جلب المخالفات من DB"""
	df = fetch_db_anomalies()
	if df is not None and len(df) > 0:
	# تحويل أسماء الأعمدة للعربية
	col_map = {
	'ID': 'الرقم', 'ContainerNumber': 'الحاوية', 'Category': 'الأصناف',
	'DescriptionEN': 'الوصف', 'RiskLevel': 'الخطورة',
	'ActualGoods': 'المحتوى', 'CreatedDate': 'التاريخ'
	}
	df = df.rename(columns={k:v for k,v in col_map.items() if k in df.columns})
	return df
	return get_anomalies()

	def generate_dashboard_html(stats):
	"""HTML لوحة التحكم مع بيانات حقيقية"""
	if stats is None:
	try:
	return stats_html()
	except:
	stats = {'total': 0, 'match': 0, 'mismatch': 0, 'high_risk': 0, 'patterns': 0, 'keywords': 0}

	src = "🟢 متصل بقاعدة البيانات" if stats.get('total', 0) > 0 else "🔴 بيانات تجريبية"

	return f"""<div style='direction:rtl;'>
	<div style='display:flex;gap:6px;flex-wrap:wrap;margin:8px 0;'>
	<div style='flex:1;min-width:120px;background:#1565C0;color:white;padding:12px;border-radius:10px;text-align:center;'>
	<div style='font-size:28px;font-weight:bold;'>{stats.get('total',0):,}</div>
	<div>📦 إجمالي الصور</div>
	</div>
	<div style='flex:1;min-width:120px;background:#2E7D32;color:white;padding:12px;border-radius:10px;text-align:center;'>
	<div style='font-size:28px;font-weight:bold;'>{stats.get('match',0):,}</div>
	<div>✅ مطابق</div>
	</div>
	<div style='flex:1;min-width:120px;background:#E65100;color:white;padding:12px;border-radius:10px;text-align:center;'>
	<div style='font-size:28px;font-weight:bold;'>{stats.get('mismatch',0):,}</div>
	<div>❌ غير مطابق</div>
	</div>
	<div style='flex:1;min-width:120px;background:#C62828;color:white;padding:12px;border-radius:10px;text-align:center;'>
	<div style='font-size:28px;font-weight:bold;'>{stats.get('high_risk',0):,}</div>
	<div>🔴 خطر عالي</div>
	</div>
	<div style='flex:1;min-width:120px;background:#6A1B9A;color:white;padding:12px;border-radius:10px;text-align:center;'>
	<div style='font-size:28px;font-weight:bold;'>{stats.get('patterns',0)}</div>
	<div>🕵️ أنماط تهريب</div>
	</div>
	<div style='flex:1;min-width:120px;background:#00695C;color:white;padding:12px;border-radius:10px;text-align:center;'>
	<div style='font-size:28px;font-weight:bold;'>{stats.get('keywords',0)}</div>
	<div>🔑 كلمات مفتاحية</div>
	</div>
	</div>
	<div style='text-align:center;padding:4px;font-size:11px;color:#666;'>{src}</div>
	</div>"""

	def analyze_image(img, declared_text):
	if img is None:
	return ("<div style='text-align:center;padding:60px;color:#999;'>"
	"<div style='font-size:48px;'>📷</div><div>ارفع صورة للبدء</div></div>",
	pd.DataFrame(), None)

	cls_result = classify_image(img)
	detected_items = cls_result['ensemble']
	n_models = cls_result['n_models']
	model_details = cls_result['models']

	annotated_img, detections = detect_objects(img) if DETECTION_MODEL else (None, [])

	# ✅ Detection الأولوية — إذا Detection اكتشف بثقة >50% يكون النتيجة الأساسية
	# ═══ Anomaly Detection ═══
	full_feat = extract_features_from_image(img)
	full_anom_score, full_is_anom, full_anom_det = check_anomaly(full_feat)
	regions = scan_container_regions(img)
	suspicious = [r for r in regions if r['is_anomaly']]
	anomaly_img = draw_arabic_boxes(img.copy(), regions) if regions else None

	final_items = []
	if detections:
	high_conf_det = [d for d in detections if d['confidence'] > 0.50 and d['name'].lower() in CARGO_DATABASE]
	if high_conf_det:
	seen = set()
	for d in sorted(high_conf_det, key=lambda x: x['confidence'], reverse=True):
	name = d['name'].lower()
	if name not in seen:
	final_items.append((name, d['confidence']))
	seen.add(name)
	else:
	final_items = detected_items
	else:
	final_items = detected_items

	total_duty = 0
	rows = []
	for i, (item, conf) in enumerate(final_items):
	info = get_hs_info(item)
	total_duty += info['duty']
	source = "🎯 Detection" if any(d['name'].lower() == item.lower() for d in detections) else "🤖 Classification"
	rows.append({'#': i+1, 'الصنف': info['ar'], 'EN': info['en'], 'كود_HS': info['hs'],
	'رمز_TSC': info['tsc_code'], 'الفصل': f"Ch.{info['ch']} {info['ch_name']}",
	'الرسوم%': info['duty'],
	'السعر_المرجعي': f"${info['avg_price']}" if info['avg_price'] > 0 else '—',
	'الثقة': f"{conf:.1%}", 'المصدر': source})
	df = pd.DataFrame(rows)

	declared = [d.strip().lower() for d in declared_text.split('+') if d.strip()] if declared_text else []
	det_names = [it.lower() for it, _ in final_items]
	if declared:
	matched = sum(1 for d in declared if any(d in nm for nm in det_names))
	is_match = matched / max(len(declared), 1) > 0.5
	else:
	is_match = random.choice([True, True, True, False])

	risk = random.randint(0, 2) if is_match else random.randint(3, 5)
	for item, _ in detected_items:
	if item == 'weapons':
	risk = 5
	break

	risk_cfg = {0:('آمن','#2E7D32','🟢'),1:('منخفض','#558B2F','🟢'),2:('متوسط','#F9A825','🟡'),
	3:('مشبوه','#EF6C00','🟠'),4:('عالي','#D84315','🔴'),5:('حرج','#B71C1C','⛔')}
	r_name, r_color, r_icon = risk_cfg[risk]

	n_items = len(final_items)
	en_desc = ' \| '.join([f"{CARGO_DATABASE.get(it,{}).get('hs','')}-{it.upper()}" for it, _ in final_items])
	ar_desc = ' \| '.join([CARGO_DATABASE.get(it,{}).get('ar', it) for it, _ in final_items])

	det_primary = detections and any(d['confidence'] > 0.80 and d['name'].lower() in CARGO_DATABASE for d in detections)
	source_text = "🎯 Detection (الأولوية)" if det_primary else "🤖 Classification"
	mode_text = f"🤖 {n_models} نماذج حقيقية \| {source_text}" if n_models > 0 else "⚙️ وضع المحاكاة"
	mode_color = "#2E7D32" if n_models > 0 else "#EF6C00"
	det_text = f" \| 📍 {len(detections)} كائنات" if detections else ""

	model_html = ""
	if model_details and n_models > 0:
	model_html = "<div style='margin-top:8px;'>"
	model_html += "<table style='width:100%;border-collapse:collapse;font-size:12px;'>"
	model_html += "<tr style='background:#1565C0;color:white;'><th style='padding:6px 8px;text-align:right;border-radius:8px 0 0 0;'>النموذج</th><th style='padding:6px;text-align:center;'>الصنف</th><th style='padding:6px;text-align:center;'>HS</th><th style='padding:6px;text-align:center;'>الرسوم</th><th style='padding:6px 8px;text-align:center;border-radius:0 8px 0 0;'>الثقة</th></tr>"
	colors = {'ConvNeXt-V2':'#1565C0','ResNet152':'#2E7D32','EfficientNet-V2':'#6A1B9A',
	'Deep-SOSUFS-v3':'#E65100','YOLOv10x-cls':'#00838F','YOLO11x-cls':'#D81B60'}
	for idx, (mn, md) in enumerate(model_details.items()):
	c = colors.get(mn, '#333')
	info = get_hs_info(md['top1'])
	bg = '#f8f9fa' if idx % 2 == 0 else '#fff'
	conf_pct = md['confidence'] * 100
	bar_color = '#2E7D32' if conf_pct > 80 else '#F9A825' if conf_pct > 50 else '#E53935'
	model_html += f"<tr style='background:{bg};border-bottom:1px solid #e0e0e0;'>"
	model_html += f"<td style='padding:5px 8px;font-weight:bold;color:{c};white-space:nowrap;'>{mn}</td>"
	model_html += f"<td style='padding:5px;text-align:center;'>{info['ar']}</td>"
	model_html += f"<td style='padding:5px;text-align:center;font-family:monospace;color:#555;'>{info['hs']}</td>"
	model_html += f"<td style='padding:5px;text-align:center;'>{info['duty']}%</td>"
	model_html += f"<td style='padding:5px 8px;text-align:center;'><div style='background:#eee;border-radius:10px;overflow:hidden;height:18px;position:relative;'><div style='background:{bar_color};height:100%;width:{conf_pct}%;border-radius:10px;'></div><span style='position:absolute;top:0;left:0;right:0;font-size:11px;font-weight:bold;line-height:18px;color:#333;'>{conf_pct:.0f}%</span></div></td>"
	model_html += "</tr>"
	model_html += "</table></div>"

	det_html = ""
	if detections:
	det_html = "<div style='margin-top:6px;display:flex;gap:6px;flex-wrap:wrap;'>"
	for d in detections:
	dinfo = get_hs_info(d['name']) if d['name'] in CARGO_DATABASE else None
	conf_pct = d['confidence'] * 100
	dc = '#E53935' if conf_pct > 70 else '#FB8C00' if conf_pct > 40 else '#43A047'
	det_html += f"<div style='background:{dc};color:white;padding:4px 10px;border-radius:15px;font-size:11px;font-weight:bold;'>{d['name'].upper()} {conf_pct:.0f}%{' · '+dinfo['hs'] if dinfo else ''}</div>"
	det_html += "</div>"

	# Anomaly HTML
	anomaly_html = ""
	if ANOMALY_MODELS:
	ac = '#E53935' if full_is_anom else '#43A047'
	ai = '🔴 مشبوه' if full_is_anom else '🟢 طبيعي'
	anomaly_html = f"<div style='background:{ac}15;border:1px solid {ac}40;border-radius:8px;padding:5px 8px;margin-top:4px;font-size:11px;'><b style='color:{ac};'>{ai} ({full_anom_score:.0%})</b>"
	if suspicious:
	for r in suspicious:
	anomaly_html += f" \| ⚠️ قسم {r['section']}: {r['label_ar']}"
	anomaly_html += "</div>"

	html = f"""
	<div style='font-family:Tajawal,sans-serif;'>
	<div style='background:{mode_color};padding:8px 14px;border-radius:10px;text-align:center;color:white;margin-bottom:8px;font-size:13px;'>{mode_text}{det_text}</div>
	<div style='background:{"#2E7D32" if is_match else "#EF6C00"};padding:14px;border-radius:12px;text-align:center;color:white;margin-bottom:10px;'>
	<div style='font-size:22px;font-weight:bold;'>{"✅ مطابق للتصريح" if is_match else "⚠️ يتطلب مراجعة"}</div>
	</div>
	<div style='display:grid;grid-template-columns:repeat(4,1fr);gap:8px;margin-bottom:10px;'>
	<div style='background:#E3F2FD;padding:12px;border-radius:10px;text-align:center;border:1px solid #90CAF9;'>
	<div style='font-size:24px;font-weight:bold;color:#1565C0;'>{n_items}</div><div style='color:#666;font-size:12px;'>أصناف</div></div>
	<div style='background:#FFF3E0;padding:12px;border-radius:10px;text-align:center;border:1px solid #FFCC80;'>
	<div style='font-size:24px;font-weight:bold;color:#E65100;'>{total_duty}%</div><div style='color:#666;font-size:12px;'>الرسوم</div></div>
	<div style='background:{r_color}15;padding:12px;border-radius:10px;text-align:center;border:1px solid {r_color}40;'>
	<div style='font-size:24px;font-weight:bold;color:{r_color};'>{r_icon} {r_name}</div><div style='color:#666;font-size:12px;'>الخطورة</div></div>
	<div style='background:#E8F5E9;padding:12px;border-radius:10px;text-align:center;border:1px solid #A5D6A7;'>
	<div style='font-size:24px;font-weight:bold;color:#2E7D32;'>TSC ✓</div><div style='color:#666;font-size:12px;'>أكواد</div></div>
	</div>
	{model_html}
	{det_html}
	{anomaly_html}
	<div style='background:#fff;padding:14px;border-radius:10px;border:1px solid #e0e0e0;margin-top:8px;margin-bottom:8px;'>
	<div style='font-weight:bold;color:#1565C0;margin-bottom:6px;font-size:13px;'>📋 HS Codes (EN)</div>
	<div style='background:#F5F5F5;padding:8px 12px;border-radius:8px;font-family:monospace;font-size:12px;direction:ltr;color:#333;'>{en_desc}</div>
	</div>
	<div style='background:#fff;padding:14px;border-radius:10px;border:1px solid #e0e0e0;'>
	<div style='font-weight:bold;color:#E65100;margin-bottom:6px;font-size:13px;'>📋 الوصف بالعربي</div>
	<div style='background:#FFF8E1;padding:8px 12px;border-radius:8px;font-size:13px;color:#333;'>{ar_desc}</div>
	</div>
	</div>"""
	output_img = anomaly_img if (suspicious or full_is_anom) and anomaly_img else annotated_img
	return html, df, output_img

	# ═══════════════════════════════════════════════════════════════
	# 📊 الإحصائيات
	# ═══════════════════════════════════════════════════════════════

	def stats_html():
	s = get_stats()
	cards = [
	(s['total'],'📦 إجمالي الحاويات','#1565C0','#1976D2'),
	(s['match'],'✅ مطابق','#2E7D32','#43A047'),
	(s['mismatch'],'❌ مخالف','#EF6C00','#FB8C00'),
	(s['high_risk'],'🔴 خطر عالي','#C62828','#E53935'),
	(f"{s['hs_codes']:,}",'🏷️ أكواد HS','#4527A0','#5E35B1'),
	(f"{s['models']}+{'1' if s['detection'] else '0'}",'🤖 نماذج AI','#00695C','#00897B'),
	]
	html = "<div style='display:grid;grid-template-columns:repeat(3,1fr);gap:10px;'>"
	for val, label, c1, c2 in cards:
	html += f"<div style='background:linear-gradient(135deg,{c1},{c2});padding:16px;border-radius:12px;text-align:center;color:white;'><div style='font-size:28px;font-weight:bold;'>{val if isinstance(val,str) else f'{val:,}'}</div><div style='opacity:0.9;font-size:12px;'>{label}</div></div>"
	html += "</div>"
	if MODELS_LOADED:
	names = ', '.join(CLASSIFICATION_MODELS.keys())
	det = "✅ Detection" if DETECTION_MODEL else "❌ Detection"
	html += f"<div style='background:#E8F5E9;padding:10px;border-radius:10px;margin-top:10px;border:1px solid #A5D6A7;text-align:center;'><b style='color:#2E7D32;'>🤖 النماذج مفعّلة:</b> <span style='font-size:12px;'>{names} \| {det}</span></div>"
	else:
	html += "<div style='background:#FFF3E0;padding:10px;border-radius:10px;margin-top:10px;border:1px solid #FFCC80;text-align:center;'><b style='color:#EF6C00;'>⚙️ وضع المحاكاة</b> <span style='font-size:12px;color:#666;'>— يحتاج GPU + torch</span></div>"
	return html

	# ═══════════════════════════════════════════════════════════════
	# 🖥️ التطبيق
	# ═══════════════════════════════════════════════════════════════

	CSS = """
	@import url('https://fonts.googleapis.com/css2?family=Tajawal:wght@400;500;700;800;900&family=Amiri:wght@400;700&display=swap');
	* { font-family: 'Tajawal', sans-serif !important; }
	.gradio-container { max-width: 1400px !important; margin: auto !important; direction: rtl !important; }
	.gr-dataframe { direction: rtl !important; font-size: 13px !important; }
	table th, table td { text-align: right !important; padding: 6px 10px !important; }
	footer { display: none !important; }
	.gr-box, .gr-form, .gr-panel { direction: rtl !important; text-align: right !important; }
	table { direction: rtl !important; }
	"""

	with gr.Blocks(title=f"SONAR-AI v{VERSION}") as app:

	# ═══ شاشة تسجيل الدخول المخصصة ═══
	with gr.Column(visible=True) as login_page:
	gr.HTML("""
	<div style='min-height:60vh;background:linear-gradient(135deg,#0a1628,#0d1f3c,#0a1628);padding:40px 20px;border-radius:16px;display:flex;align-items:center;justify-content:center;font-family:Tajawal,sans-serif;direction:rtl;'>
	<div style='background:rgba(255,255,255,0.03);border:1px solid rgba(212,168,67,0.2);border-radius:20px;padding:40px;width:380px;max-width:90%;text-align:center;backdrop-filter:blur(10px);box-shadow:0 8px 32px rgba(0,0,0,0.4);'>
	<div style='font-size:60px;margin-bottom:10px;'>🔱</div>
	<div style='font-size:36px;font-weight:900;background:linear-gradient(135deg,#f0d68a,#d4a843,#c49032);-webkit-background-clip:text;-webkit-text-fill-color:transparent;letter-spacing:4px;'>SONAR-AI</div>
	<div style='color:#4ecdc4;font-size:13px;letter-spacing:2px;margin-bottom:5px;'>نظام الفحص الذكي بالذكاء الاصطناعي</div>
	<div style='color:rgba(240,214,138,0.6);font-size:15px;margin-bottom:25px;'>الهيئة العامة للكمارك العراقية</div>
	</div>
	</div>
	""")
	with gr.Row():
	gr.Column(scale=1)
	with gr.Column(scale=2):
	login_user = gr.Textbox(label="👤 اسم المستخدم", value="عباس", text_align="right")
	login_pass = gr.Textbox(label="🔑 كلمة المرور", type="password", value="1", text_align="right")
	login_btn = gr.Button("🔱 تسجيل الدخول", variant="primary", size="lg")
	login_msg = gr.HTML("")
	gr.Column(scale=1)

	# ═══ التطبيق الرئيسي ═══
	with gr.Column(visible=False) as main_page:
	with gr.Row():
	gr.HTML(f"""
	<div style='background:linear-gradient(135deg,#0D47A1,#1565C0,#1976D2);padding:16px 20px;border-radius:14px;text-align:center;color:white;margin-bottom:10px;flex:1;'>
	<h1 style='margin:0;font-size:26px;font-weight:800;'>🔱 SONAR-AI v{VERSION}</h1>
	<p style='margin:3px 0 0;font-size:13px;opacity:0.9;'>{len(CLASSIFICATION_MODELS)} Classification + {'Detection' if DETECTION_MODEL else 'Sim'} \| {len(TSC_DATABASE):,} HS \| 8 Physics (244D)</p>
	</div>""")
	exit_btn = gr.Button("🚪 خروج", variant="stop", size="sm", scale=0, min_width=80)

	with gr.Tabs():
	with gr.Tab("📊 لوحة التحكم", id="dashboard"):
	d_stats = gr.HTML(generate_dashboard_html(fetch_db_stats()))
	d_refresh = gr.Button("🔄 تحديث من قاعدة البيانات", variant="primary", size="lg")
	gr.Markdown("### 📋 آخر الحاويات")
	d_tbl = gr.Dataframe(value=get_anomalies())
	d_refresh.click(refresh_dashboard, outputs=[d_stats, d_tbl])

	with gr.Tab("🔬 تحليل الصور"):
	gr.HTML("<div style='background:#E3F2FD;padding:10px 14px;border-radius:10px;margin-bottom:8px;border-right:4px solid #1565C0;'><b style='color:#1565C0;'>📷 Classification + Detection + HS Codes</b></div>")
	with gr.Row():
	with gr.Column(scale=1):
	a_img = gr.Image(label="📷 صورة الأشعة", type="pil", height=200)
	gr.HTML("""<div style='background:linear-gradient(135deg,#0d47a1,#1565C0,#1976D2);padding:5px 8px;border-radius:8px;margin:4px 0;display:flex;gap:4px;flex-wrap:wrap;justify-content:center;align-items:center;'>
	<span style='color:#ffd740;font-weight:bold;font-size:11px;margin-left:6px;'>🎛️ أدوات السونار</span>
	<button onclick="document.querySelector('#btn_gray').click()" style='background:#607D8B;color:white;border:none;padding:4px 10px;border-radius:6px;cursor:pointer;font-size:11px;font-weight:bold;'>⬜ رمادي</button>
	<button onclick="document.querySelector('#btn_inv').click()" style='background:#455A64;color:white;border:none;padding:4px 10px;border-radius:6px;cursor:pointer;font-size:11px;font-weight:bold;'>◐ عكس</button>
	<button onclick="document.querySelector('#btn_thermal').click()" style='background:#E53935;color:white;border:none;padding:4px 10px;border-radius:6px;cursor:pointer;font-size:11px;font-weight:bold;'>🔥 حراري</button>
	<button onclick="document.querySelector('#btn_cool').click()" style='background:#1E88E5;color:white;border:none;padding:4px 10px;border-radius:6px;cursor:pointer;font-size:11px;font-weight:bold;'>❄ بارد</button>
	<button onclick="document.querySelector('#btn_rainbow').click()" style='background:#7B1FA2;color:white;border:none;padding:4px 10px;border-radius:6px;cursor:pointer;font-size:11px;font-weight:bold;'>🌈 ألوان</button>
	<button onclick="document.querySelector('#btn_edge').click()" style='background:#F57C00;color:white;border:none;padding:4px 10px;border-radius:6px;cursor:pointer;font-size:11px;font-weight:bold;'>📐 حواف</button>
	<button onclick="document.querySelector('#btn_contrast').click()" style='background:#00897B;color:white;border:none;padding:4px 10px;border-radius:6px;cursor:pointer;font-size:11px;font-weight:bold;'>🔆 تباين</button>
	<button onclick="document.querySelector('#btn_bright').click()" style='background:#FDD835;color:#333;border:none;padding:4px 10px;border-radius:6px;cursor:pointer;font-size:11px;font-weight:bold;'>☀ سطوع</button>
	</div>""")
	with gr.Row(visible=False):
	btn_gray = gr.Button("⬜", elem_id="btn_gray", size="sm", min_width=30)
	btn_inv = gr.Button("◐", elem_id="btn_inv", size="sm", min_width=30)
	btn_thermal = gr.Button("🔥", elem_id="btn_thermal", size="sm", min_width=30)
	btn_cool = gr.Button("❄", elem_id="btn_cool", size="sm", min_width=30)
	btn_rainbow = gr.Button("🌈", elem_id="btn_rainbow", size="sm", min_width=30)
	btn_edge = gr.Button("📐", elem_id="btn_edge", size="sm", min_width=30)
	btn_contrast = gr.Button("🔆", elem_id="btn_contrast", size="sm", min_width=30)
	btn_bright = gr.Button("☀", elem_id="btn_bright", size="sm", min_width=30)
	a_dec = gr.Textbox(label="📋 البضاعة المصرّح بها", placeholder="SHOES + FABRIC + ...", lines=1)
	a_btn = gr.Button("🔍 تحليل شامل", variant="primary", size="lg")
	with gr.Column(scale=1):
	a_res = gr.HTML("<div style='text-align:center;padding:30px;color:#999;direction:rtl;'>📷 ارفع صورة للبدء</div>")
	a_det = gr.Image(label="📍 Detection / Anomaly", type="pil", height=200)
	a_tbl = gr.Dataframe(label="📋 الأصناف + HS", value=pd.DataFrame(), wrap=True)
	a_btn.click(analyze_image, inputs=[a_img, a_dec], outputs=[a_res, a_tbl, a_det], api_name="analyze")
	btn_gray.click(apply_grayscale, inputs=[a_img], outputs=[a_img])
	btn_inv.click(apply_invert, inputs=[a_img], outputs=[a_img])
	btn_thermal.click(apply_thermal, inputs=[a_img], outputs=[a_img])
	btn_cool.click(apply_cool, inputs=[a_img], outputs=[a_img])
	btn_rainbow.click(apply_rainbow, inputs=[a_img], outputs=[a_img])
	btn_edge.click(apply_edge, inputs=[a_img], outputs=[a_img])
	btn_contrast.click(apply_contrast, inputs=[a_img], outputs=[a_img])
	btn_bright.click(apply_brightness, inputs=[a_img], outputs=[a_img])

	# ═══ 🆕 التحليل الفيزيائي ═══
	with gr.Tab("🔬 التحليل الفيزيائي 🆕"):
	gr.HTML("""<div style='background:linear-gradient(135deg,#4A148C,#6A1B9A);padding:12px 14px;border-radius:10px;margin-bottom:8px;color:white;text-align:center;'>
	<b>🔬 8 تقنيات فيزيائية مبتكرة — 244 بُعد — كشف الإخفاء</b><br>
	<span style='font-size:11px;opacity:0.8;'>DPM · CKB · 3DTW · CWS · TCV · DWE · MVF · ENV \| Dr. Abbas Fadel Al-Jubouri</span>
	</div>""")
	with gr.Row():
	with gr.Column(scale=1):
	p_img = gr.Image(label="📷 صورة الأشعة", type="pil", height=250)
	p_btn = gr.Button("🔬 تحليل فيزيائي شامل", variant="primary", size="lg")
	gr.HTML("""<div style='background:#F3E5F5;padding:10px;border-radius:8px;border:1px solid #CE93D8;font-size:11px;color:#4A148C;margin-top:6px;'>
	<b>8 تقنيات:</b> 🔥DPM(4D) · 📋CKB(8D) · 🌊3DTW(25D) · 🌀CWS(50D)<br>
	🌪️TCV(42D) · 🔦DWE(35D) · 🌋MVF(40D) · 🏜️ENV(40D) = <b>244D</b>
	</div>""")
	with gr.Column(scale=2):
	p_res = gr.HTML("<div style='text-align:center;padding:60px;color:#999;direction:rtl;'><div style='font-size:48px;'>🔬</div><div>ارفع صورة X-Ray للتحليل الفيزيائي</div><div style='font-size:12px;margin-top:10px;'>8 تقنيات — 244 بُعد — كشف الإخفاء</div></div>")
	p_tbl = gr.Dataframe(label="📊 نتائج التقنيات الثمانية", value=pd.DataFrame(), wrap=True)
	p_btn.click(physics_analyze, inputs=[p_img], outputs=[p_res, p_tbl])


	with gr.Tab("🏷️ قاعدة الأصناف"):
	gr.HTML(f"<div style='background:#E8F5E9;padding:10px 14px;border-radius:10px;margin-bottom:8px;border-right:4px solid #2E7D32;'><b style='color:#2E7D32;'>📦 {len(CARGO_DATABASE)} صنف</b></div>")
	cats = gr.Dataframe(value=get_categories())
	gr.Button("🔄", size="sm").click(get_categories, outputs=cats)

	with gr.Tab("🔍 بحث TSC"):
	gr.HTML(f"<div style='background:#FFF3E0;padding:10px 14px;border-radius:10px;margin-bottom:8px;border-right:4px solid #EF6C00;'><b style='color:#EF6C00;'>🔍 TSC — {len(TSC_DATABASE):,} كود</b></div>")
	with gr.Row():
	tsc_q = gr.Textbox(label="بحث", placeholder="870323 أو أحذية ...", scale=3)
	tsc_btn = gr.Button("🔍", variant="primary", scale=1)
	tsc_res = gr.Dataframe(value=pd.DataFrame())
	tsc_btn.click(search_tsc, inputs=[tsc_q], outputs=[tsc_res])
	tsc_q.submit(search_tsc, inputs=[tsc_q], outputs=[tsc_res])

	with gr.Tab("👥 فريق العمل"):
	gr.HTML("""
	<div style='min-height:80vh;background:linear-gradient(135deg,#0a1628,#0d1f3c,#0a1628);padding:30px 20px;color:#f0f0f0;text-align:center;border-radius:16px;font-family:Tajawal,sans-serif;direction:rtl;'>
	<div style='font-size:50px;'>🔱</div>
	<div style='font-size:36px;font-weight:900;background:linear-gradient(135deg,#f0d68a,#d4a843,#c49032);-webkit-background-clip:text;-webkit-text-fill-color:transparent;letter-spacing:4px;'>SONAR-AI</div>
	<div style='color:#4ecdc4;font-size:12px;letter-spacing:3px;'>نظام الفحص الذكي بالذكاء الاصطناعي</div>
	<div style='color:rgba(240,214,138,0.6);font-size:16px;font-family:Amiri,serif;margin:8px 0 20px;'>الهيئة العامة للكمارك العراقية</div>
	<div style='height:1px;background:linear-gradient(90deg,transparent,#d4a843,transparent);opacity:0.4;margin:20px auto;max-width:600px;'></div>
	<div style='font-size:26px;font-family:Amiri,serif;color:#f0d68a;margin-bottom:20px;'>فريق العمل</div>
	<div style='background:linear-gradient(135deg,rgba(212,168,67,0.1),rgba(46,107,189,0.06));border:1px solid rgba(212,168,67,0.25);border-radius:16px;padding:24px;max-width:400px;margin:0 auto 20px;'>
	<div style='width:70px;height:70px;border-radius:50%;background:linear-gradient(135deg,#d4a843,#c49032);display:flex;align-items:center;justify-content:center;margin:0 auto 10px;font-size:32px;box-shadow:0 4px 15px rgba(212,168,67,0.3);'>👨‍💼</div>
	<div style='font-size:22px;font-weight:800;color:#f0d68a;'>د. عباس فاضل</div>
	<div style='font-size:13px;color:#4ecdc4;letter-spacing:2px;'>رئيس الفريق</div>
	</div>
	<div style='background:rgba(78,205,196,0.06);border:1px solid rgba(78,205,196,0.15);border-radius:12px;padding:14px 20px;max-width:400px;margin:0 auto 20px;'>
	<div style='font-size:17px;font-weight:700;color:#f0d68a;'>أ. عامر</div>
	<div style='font-size:12px;color:rgba(240,240,240,0.5);'>المشاور القانوني</div>
	</div>
	<div style='color:rgba(240,214,138,0.5);font-size:14px;letter-spacing:3px;margin-bottom:15px;'>— المدربون —</div>
	<div style='display:flex;flex-wrap:wrap;justify-content:center;gap:10px;max-width:700px;margin:0 auto 20px;'>
	<div style='background:rgba(255,255,255,0.03);border:1px solid rgba(255,255,255,0.07);border-radius:12px;padding:14px 18px;min-width:140px;'><div style='font-size:24px;'>🎓</div><div style='font-size:14px;font-weight:700;color:#f0d68a;'>أ. ناظم</div><div style='font-size:10px;color:rgba(240,240,240,0.35);'>مدرب</div></div>
	<div style='background:rgba(255,255,255,0.03);border:1px solid rgba(255,255,255,0.07);border-radius:12px;padding:14px 18px;min-width:140px;'><div style='font-size:24px;'>🎓</div><div style='font-size:14px;font-weight:700;color:#f0d68a;'>أ. ظفار</div><div style='font-size:10px;color:rgba(240,240,240,0.35);'>مدرب</div></div>
	<div style='background:rgba(255,255,255,0.03);border:1px solid rgba(255,255,255,0.07);border-radius:12px;padding:14px 18px;min-width:140px;'><div style='font-size:24px;'>🎓</div><div style='font-size:14px;font-weight:700;color:#f0d68a;'>أ. عمار الشعلان</div><div style='font-size:10px;color:rgba(240,240,240,0.35);'>مدرب</div></div>
	<div style='background:rgba(255,255,255,0.03);border:1px solid rgba(255,255,255,0.07);border-radius:12px;padding:14px 18px;min-width:140px;'><div style='font-size:24px;'>🎓</div><div style='font-size:14px;font-weight:700;color:#f0d68a;'>أ. يونس ذنون</div><div style='font-size:10px;color:rgba(240,240,240,0.35);'>مدرب</div></div>
	<div style='background:rgba(255,255,255,0.03);border:1px solid rgba(255,255,255,0.07);border-radius:12px;padding:14px 18px;min-width:140px;'><div style='font-size:24px;'>⚙️</div><div style='font-size:14px;font-weight:700;color:#f0d68a;'>م. باسم محمد جابر</div><div style='font-size:10px;color:rgba(240,240,240,0.35);'>مهندس</div></div>
	</div>
	<div style='height:1px;background:linear-gradient(90deg,transparent,#d4a843,transparent);opacity:0.4;margin:20px auto;max-width:600px;'></div>
	<div style='color:rgba(240,240,240,0.2);font-size:10px;letter-spacing:2px;'>SONAR-AI v16.8 — Powered by DL + Physics by Deep Learning</div>
	</div>
	""")

	with gr.Tab("⚡ الخطورة"):
	gr.HTML("""<div style='display:flex;gap:6px;margin:12px 0;flex-wrap:wrap;'>
	<span style='background:#2E7D32;color:white;padding:8px 16px;border-radius:20px;'>🟢 0-آمن</span>
	<span style='background:#558B2F;color:white;padding:8px 16px;border-radius:20px;'>🟢 1-منخفض</span>
	<span style='background:#F9A825;color:#333;padding:8px 16px;border-radius:20px;'>🟡 2-متوسط</span>
	<span style='background:#EF6C00;color:white;padding:8px 16px;border-radius:20px;'>🟠 3-مشبوه</span>
	<span style='background:#D84315;color:white;padding:8px 16px;border-radius:20px;'>🔴 4-عالي</span>
	<span style='background:#B71C1C;color:white;padding:8px 16px;border-radius:20px;'>⛔ 5-حرج</span></div>""")
	gr.Markdown("""### 📋 معايير الخطورة
	\| المستوى \| الحالة \| الإجراء \|
	\|---------\|--------\|---------\|
	\| 0-آمن \| مطابقة كاملة \| تمرير \|
	\| 1-منخفض \| اختلاف بسيط \| مراجعة وثائق \|
	\| 2-متوسط \| صنف إضافي \| فحص عشوائي \|
	\| 3-مشبوه \| عدة مخالفات \| تفتيش دقيق \|
	\| 4-عالي \| أصناف مقيّدة \| تفتيش شامل \|
	\| 5-حرج \| ممنوعات/أسلحة \| إيقاف فوري \|""")

	gr.HTML(f"<div style='text-align:center;padding:10px;color:#999;font-size:11px;'>🔱 SONAR-AI v{VERSION} \| {len(CLASSIFICATION_MODELS)} Models \| {len(TSC_DATABASE):,} HS \| © 2026</div>")

	# ═══ دالة تسجيل الدخول ═══
	def do_login(username, password):
	if authenticate(username, password):
	return gr.update(visible=False), gr.update(visible=True), ""
	return gr.update(visible=True), gr.update(visible=False), "<div style='color:#ef5350;text-align:center;font-weight:bold;padding:8px;'>❌ اسم المستخدم أو كلمة المرور غير صحيحة</div>"

	# ═══ شاشة الخروج (فريق العمل) ═══
	with gr.Column(visible=False) as exit_page:
	gr.HTML("""
	<div style='min-height:85vh;background:linear-gradient(135deg,#0a1628,#0d1f3c,#0a1628);padding:30px 20px;color:#f0f0f0;text-align:center;border-radius:16px;font-family:Tajawal,sans-serif;direction:rtl;'>
	<div style='font-size:50px;'>🔱</div>
	<div style='font-size:36px;font-weight:900;background:linear-gradient(135deg,#f0d68a,#d4a843,#c49032);-webkit-background-clip:text;-webkit-text-fill-color:transparent;letter-spacing:4px;'>SONAR-AI</div>
	<div style='color:#4ecdc4;font-size:12px;letter-spacing:3px;'>نظام الفحص الذكي بالذكاء الاصطناعي</div>
	<div style='color:rgba(240,214,138,0.6);font-size:16px;margin:8px 0 20px;'>الهيئة العامة للكمارك العراقية</div>
	<div style='height:1px;background:linear-gradient(90deg,transparent,#d4a843,transparent);opacity:0.4;margin:20px auto;max-width:600px;'></div>
	<div style='font-size:28px;color:#f0d68a;margin-bottom:20px;'>فريق العمل</div>
	<div style='background:linear-gradient(135deg,rgba(212,168,67,0.1),rgba(46,107,189,0.06));border:1px solid rgba(212,168,67,0.25);border-radius:16px;padding:24px;max-width:400px;margin:0 auto 20px;'>
	<div style='width:70px;height:70px;border-radius:50%;background:linear-gradient(135deg,#d4a843,#c49032);display:flex;align-items:center;justify-content:center;margin:0 auto 10px;font-size:32px;box-shadow:0 4px 15px rgba(212,168,67,0.3);'>👨‍💼</div>
	<div style='font-size:22px;font-weight:800;color:#f0d68a;'>د. عباس فاضل</div>
	<div style='font-size:13px;color:#4ecdc4;letter-spacing:2px;'>رئيس الفريق</div>
	</div>
	<div style='background:rgba(78,205,196,0.06);border:1px solid rgba(78,205,196,0.15);border-radius:12px;padding:14px 20px;max-width:400px;margin:0 auto 20px;'>
	<div style='font-size:17px;font-weight:700;color:#f0d68a;'>أ. عامر</div>
	<div style='font-size:12px;color:rgba(240,240,240,0.5);'>المشاور القانوني</div>
	</div>
	<div style='color:rgba(240,214,138,0.5);font-size:14px;letter-spacing:3px;margin-bottom:15px;'>— المدربون —</div>
	<div style='display:flex;flex-wrap:wrap;justify-content:center;gap:10px;max-width:700px;margin:0 auto 20px;'>
	<div style='background:rgba(255,255,255,0.03);border:1px solid rgba(255,255,255,0.07);border-radius:12px;padding:14px 18px;min-width:140px;'><div style='font-size:24px;'>🎓</div><div style='font-size:14px;font-weight:700;color:#f0d68a;'>أ. ناظم</div><div style='font-size:10px;color:rgba(240,240,240,0.35);'>مدرب</div></div>
	<div style='background:rgba(255,255,255,0.03);border:1px solid rgba(255,255,255,0.07);border-radius:12px;padding:14px 18px;min-width:140px;'><div style='font-size:24px;'>🎓</div><div style='font-size:14px;font-weight:700;color:#f0d68a;'>أ. ظفار</div><div style='font-size:10px;color:rgba(240,240,240,0.35);'>مدرب</div></div>
	<div style='background:rgba(255,255,255,0.03);border:1px solid rgba(255,255,255,0.07);border-radius:12px;padding:14px 18px;min-width:140px;'><div style='font-size:24px;'>🎓</div><div style='font-size:14px;font-weight:700;color:#f0d68a;'>أ. عمار الشعلان</div><div style='font-size:10px;color:rgba(240,240,240,0.35);'>مدرب</div></div>
	<div style='background:rgba(255,255,255,0.03);border:1px solid rgba(255,255,255,0.07);border-radius:12px;padding:14px 18px;min-width:140px;'><div style='font-size:24px;'>🎓</div><div style='font-size:14px;font-weight:700;color:#f0d68a;'>أ. يونس ذنون</div><div style='font-size:10px;color:rgba(240,240,240,0.35);'>مدرب</div></div>
	<div style='background:rgba(255,255,255,0.03);border:1px solid rgba(255,255,255,0.07);border-radius:12px;padding:14px 18px;min-width:140px;'><div style='font-size:24px;'>⚙️</div><div style='font-size:14px;font-weight:700;color:#f0d68a;'>م. باسم محمد جابر</div><div style='font-size:10px;color:rgba(240,240,240,0.35);'>مهندس</div></div>
	</div>
	<div style='height:1px;background:linear-gradient(90deg,transparent,#d4a843,transparent);opacity:0.4;margin:20px auto;max-width:600px;'></div>
	<div style='color:rgba(240,240,240,0.3);font-size:11px;margin-top:10px;'>شكراً لاستخدامكم SONAR-AI v16.8</div>
	</div>
	""")
	relogin_btn = gr.Button("🔙 العودة لتسجيل الدخول", variant="secondary", size="lg")

	# ═══ أحداث تسجيل الدخول ═══
	def do_login(username, password):
	if authenticate(username, password):
	return gr.update(visible=False), gr.update(visible=True), gr.update(visible=False), ""
	return gr.update(visible=True), gr.update(visible=False), gr.update(visible=False), "<div style='color:#ef5350;text-align:center;font-weight:bold;padding:8px;'>❌ اسم المستخدم أو كلمة المرور غير صحيحة</div>"

	def do_exit():
	return gr.update(visible=False), gr.update(visible=True)

	def do_relogin():
	return gr.update(visible=True), gr.update(visible=False), gr.update(visible=False)

	login_btn.click(do_login, inputs=[login_user, login_pass], outputs=[login_page, main_page, exit_page, login_msg])
	login_pass.submit(do_login, inputs=[login_user, login_pass], outputs=[login_page, main_page, exit_page, login_msg])
	exit_btn.click(do_exit, outputs=[main_page, exit_page])
	relogin_btn.click(do_relogin, outputs=[login_page, main_page, exit_page])

	# ═══════════════════════════════════════════════════════════════
	# 🔌 API Endpoint for C# Desktop App (No Auth Required)
	# ═══════════════════════════════════════════════════════════════

	def api_analyze(img, declared_text=""):
	"""API endpoint — returns JSON results for C# integration"""
	if img is None:
	return json.dumps({"success": False, "error": "No image provided"}, ensure_ascii=False)

	cls_result = classify_image(img)
	detected_items = cls_result['ensemble']
	n_models = cls_result['n_models']
	model_details = cls_result['models']

	_, detections = detect_objects(img) if DETECTION_MODEL else (None, [])

	# ═══ Anomaly Detection ═══
	full_feat = extract_features_from_image(img)
	full_anom_score, full_is_anom, full_anom_det = check_anomaly(full_feat)
	regions = scan_container_regions(img)
	suspicious = [r for r in regions if r['is_anomaly']]
	anomaly_img = draw_arabic_boxes(img.copy(), regions) if regions else None

	final_items = []
	source = "classification"
	if detections:
	high_conf_det = [d for d in detections if d['confidence'] > 0.50 and d['name'].lower() in CARGO_DATABASE]
	if high_conf_det:
	source = "detection"
	seen = set()
	for d in sorted(high_conf_det, key=lambda x: x['confidence'], reverse=True):
	name = d['name'].lower()
	if name not in seen:
	final_items.append((name, d['confidence']))
	seen.add(name)
	else:
	final_items = detected_items
	else:
	final_items = detected_items

	items_json = []
	total_duty = 0
	for item, conf in final_items:
	info = get_hs_info(item)
	total_duty += info['duty']
	items_json.append({
	"name_en": info['en'],
	"name_ar": info['ar'],
	"hs_code": info['hs'],
	"chapter": info['ch'],
	"chapter_name": info['ch_name'],
	"duty_percent": info['duty'],
	"avg_price": info['avg_price'],
	"tsc_code": info['tsc_code'],
	"confidence": round(conf, 4),
	"source": "detection" if any(d['name'].lower() == item.lower() for d in detections) else "classification"
	})

	cls_details = {}
	for mn, md in model_details.items():
	cls_details[mn] = {"top1": md['top1'], "confidence": round(md['confidence'], 4)}

	det_details = [{"name": d['name'], "confidence": round(d['confidence'], 4)} for d in detections]

	result = {
	"success": True,
	"version": VERSION,
	"source": source,
	"n_models": n_models,
	"total_duty": total_duty,
	"items": items_json,
	"classification": cls_details,
	"detection": det_details,
	"timestamp": datetime.now().isoformat()
	}

	return json.dumps(result, ensure_ascii=False)

	if __name__ == "__main__":
	print(f"🔱 SONAR-AI v{VERSION}")
	print(f"📦 Cargo: {len(CARGO_DATABASE)} \| TSC: {len(TSC_DATABASE)}")
	print(f"🤖 Models: {len(CLASSIFICATION_MODELS)} cls + {'det' if DETECTION_MODEL else 'no det'}")
	app.launch(
	server_name="0.0.0.0",
	server_port=7860,
	ssr_mode=False,
	theme=gr.themes.Soft(),
	css=CSS,
	show_error=True,
	)