import cv2 import numpy as np from PIL import Image import torch from torchvision import models, transforms from ultralytics import YOLO import gradio as gr import torch.nn as nn import os from datetime import datetime import json # ============================================ # RICE ANALYZER PRO - Advanced Analytics Dashboard # A professional grain analysis platform with statistics and batch processing # ============================================ # --- SYSTEM CONFIGURATION --- device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') # Model initialization try: detection_model = YOLO('best.pt') classifier_network = models.resnet50(weights=None) classifier_network.fc = nn.Linear(classifier_network.fc.in_features, 3) classifier_network.load_state_dict(torch.load('rice_resnet_model.pth', map_location=device)) classifier_network = classifier_network.to(device) classifier_network.eval() models_loaded = True except Exception as e: print(f" Model initialization failed: {e}") detection_model = None classifier_network = None models_loaded = False # Variety definitions VARIETY_MAP = { 0: "C9 Premium", 1: "Kant Special", 2: "Superfine Grade" } VARIETY_COLORS = { "C9 Premium": (255, 100, 100), # Red "Kant Special": (100, 100, 255), # Blue "Superfine Grade": (100, 255, 100) # Green } # Data preprocessing pipeline image_preprocessor = transforms.Compose([ transforms.Resize((224, 224)), transforms.ToTensor(), transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]) ]) # --- ANALYTICS FUNCTIONS --- def analyze_single_grain(grain_image): """Perform classification on an individual grain""" if not models_loaded: return "System Error" tensor_input = image_preprocessor(grain_image).unsqueeze(0).to(device) with torch.no_grad(): prediction = classifier_network(tensor_input) class_idx = torch.argmax(prediction, dim=1).item() return VARIETY_MAP[class_idx] def compute_distribution_stats(variety_counts): """Generate statistical summary of grain distribution""" total = sum(variety_counts.values()) if total == 0: return "No grains detected for analysis." stats = [" **Distribution Analysis**\n"] stats.append(f" Total Grains Detected: **{total}**\n") stats.append("\n**Breakdown by Variety:**\n") for variety, count in sorted(variety_counts.items(), key=lambda x: x[1], reverse=True): percentage = (count / total) * 100 bar_length = int(percentage / 5) bar = "█" * bar_length + "░" * (20 - bar_length) stats.append(f"- {variety}: {count} grains ({percentage:.1f}%) {bar}\n") # Dominant variety dominant = max(variety_counts.items(), key=lambda x: x[1]) stats.append(f"\n **Dominant Variety:** {dominant[0]}") return "".join(stats) def execute_batch_analysis(input_img): """Main processing pipeline with analytics""" if not models_loaded: raise gr.Error(" Analysis engine unavailable. Please check model files.") if input_img is None: raise gr.Error(" Please upload an image to begin analysis.") # Convert and prepare image img_array = np.array(input_img) img_bgr = cv2.cvtColor(img_array, cv2.COLOR_RGB2BGR) # Detection phase detection_results = detection_model(img_bgr, verbose=False)[0] bounding_boxes = detection_results.boxes.xyxy.cpu().numpy() if len(bounding_boxes) == 0: gr.Warning("⚠️ No rice grains found. Try a clearer image.") return ( Image.fromarray(cv2.cvtColor(img_bgr, cv2.COLOR_BGR2RGB)), "No grains detected in the provided image." ) # Classification phase with tracking variety_counter = {v: 0 for v in VARIETY_MAP.values()} for box in bounding_boxes: x1, y1, x2, y2 = map(int, box[:4]) grain_crop = img_bgr[y1:y2, x1:x2] if grain_crop.shape[0] > 0 and grain_crop.shape[1] > 0: grain_pil = Image.fromarray(cv2.cvtColor(grain_crop, cv2.COLOR_BGR2RGB)) variety_label = analyze_single_grain(grain_pil) variety_counter[variety_label] += 1 # Visualization with color coding color = VARIETY_COLORS[variety_label] cv2.rectangle(img_bgr, (x1, y1), (x2, y2), color, 3) # Label with background label_text = variety_label (text_w, text_h), _ = cv2.getTextSize(label_text, cv2.FONT_HERSHEY_SIMPLEX, 0.7, 2) cv2.rectangle(img_bgr, (x1, y1-text_h-10), (x1+text_w, y1), color, -1) cv2.putText(img_bgr, label_text, (x1, y1-5), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (255, 255, 255), 2) # Generate analytics statistics_report = compute_distribution_stats(variety_counter) return ( Image.fromarray(cv2.cvtColor(img_bgr, cv2.COLOR_BGR2RGB)), statistics_report ) # --- GRADIO INTERFACE --- custom_css = """ .gradio-container { font-family: 'Segoe UI', Tahoma, Geneva, Verdana, sans-serif; } .header-box { background: linear-gradient(135deg, #667eea 0%, #764ba2 100%); padding: 30px; border-radius: 15px; color: white; text-align: center; margin-bottom: 20px; } .stat-box { background: #f8f9fa; border-left: 4px solid #667eea; padding: 15px; border-radius: 8px; margin: 10px 0; } """ with gr.Blocks(css=custom_css, title="Rice Analyzer Pro") as app: gr.HTML("""

🌾 Rice Analyzer Pro

Advanced Grain Analytics Platform | Professional Quality Assessment

""") with gr.Tabs(): # Main Analysis Tab with gr.Tab("Analysis app"): gr.Markdown(""" ### How It Works 1. **Upload** your rice sample image (clear photos work best) 2. **Analyze** Click on "Start Analysis" and let Our AI model detect and classify each grain 3. **Review** detailed statistics and visual results **Color Coding:** 🔵 C9 Premium | 🔴 Kant Special | 🟢 Superfine Grade """) with gr.Row(): with gr.Column(scale=1): image_upload = gr.Image(type="pil", label="📸 Sample Image") analyze_button = gr.Button(" Start Analysis", variant="primary", size="lg") with gr.Column(scale=1): output_visualization = gr.Image(label=" Annotated Results") with gr.Row(): statistics_output = gr.Markdown(label=" Statistical Report") analyze_button.click( fn=execute_batch_analysis, inputs=image_upload, outputs=[output_visualization, statistics_output] ) # Documentation Tab with gr.Tab(" Documentation"): gr.Markdown(""" ## System Overview **Rice Analyzer Pro** uses a two-stage deep learning pipeline: - **Stage 1:** YOLO-based grain detection - **Stage 2:** ResNet50 variety classification ### Supported Varieties | Variety | Description | Market Grade | |---------|-------------|--------------| | C9 Premium | High-quality long grain | Premium | | Kant Special | Medium grain specialty | Standard | | Superfine Grade | Ultra-refined grain | Super Fine | ### Best Practices - Use well-lit images with minimal shadows - Ensure grains are separated (not clumped) - Plain backgrounds yield better results - Recommended resolution: 1024x1024 or higher ### Technical Specifications - Detection Model: YOLOv8 - Classification: ResNet50 (Fine-tuned) - Processing: GPU-accelerated (when available) """) gr.Markdown("---") gr.Markdown("### Sample Gallery") gr.Examples( examples=[ "samples/rice1.jpg", "samples/rice2.jpg", "samples/rice4.jpg", "samples/rice5.jpg" ], inputs=image_upload, outputs=[output_visualization, statistics_output], fn=execute_batch_analysis, label="Click any sample to run instant analysis" ) if __name__ == "__main__": app.queue() app.launch()