"""
Fallback detection module for when the main YOLO model fails due to
torchvision or other dependency issues.

This provides a simple detection mechanism without dependencies on 
PyTorch or torchvision.
"""

import logging
import numpy as np
import os
import tempfile
from typing import Dict, List, Optional, Tuple, Union
import uuid

# Initialize logger
logger = logging.getLogger(__name__)

# Try to import OpenCV, but don't fail if not available
try:
    import cv2
    HAS_CV2 = True
except ImportError:
    HAS_CV2 = False
    logger.warning("OpenCV (cv2) not available in fallback_detection module")

# Basic color detection thresholds
# These are simple HSV thresholds for detecting common pollution colors
COLOR_THRESHOLDS = {
    "oil_spill": {
        "lower": np.array([0, 0, 0]),
        "upper": np.array([180, 255, 80]),
        "label": "Potential Oil Spill",
        "confidence": 0.6
    },
    "plastic_bright": {
        "lower": np.array([0, 50, 180]),
        "upper": np.array([30, 255, 255]),
        "label": "Potential Plastic Debris",
        "confidence": 0.7
    },
    "foam_pollution": {
        "lower": np.array([0, 0, 200]),
        "upper": np.array([180, 30, 255]),
        "label": "Potential Foam/Chemical Pollution",
        "confidence": 0.65
    },
    # Enhanced plastic bottle detection thresholds
    "plastic_bottles_clear": {
        "lower": np.array([0, 0, 140]),
        "upper": np.array([180, 60, 255]),
        "label": "plastic bottle",  # Updated label to match YOLO naming
        "confidence": 0.80
    },
    "plastic_bottles_blue": {
        "lower": np.array([90, 40, 100]),
        "upper": np.array([130, 255, 255]),
        "label": "plastic bottle",
        "confidence": 0.75
    },
    "plastic_bottles_green": {
        "lower": np.array([35, 40, 100]),
        "upper": np.array([85, 255, 255]),
        "label": "plastic bottle",
        "confidence": 0.75
    },
    "plastic_bottles_white": {
        "lower": np.array([0, 0, 180]),
        "upper": np.array([180, 30, 255]),
        "label": "plastic bottle",
        "confidence": 0.75
    },
    "plastic_bottles_cap": {
        "lower": np.array([100, 100, 100]),
        "upper": np.array([140, 255, 255]),
        "label": "plastic bottle cap",
        "confidence": 0.85
    },
    "blue_plastic": {
        "lower": np.array([90, 50, 50]),
        "upper": np.array([130, 255, 255]),
        "label": "plastic waste",  # Updated label for consistency
        "confidence": 0.6
    },
    "green_plastic": {
        "lower": np.array([35, 50, 50]),
        "upper": np.array([85, 255, 255]),
        "label": "plastic waste",
        "confidence": 0.6
    },
    "white_plastic": {
        "lower": np.array([0, 0, 190]),
        "upper": np.array([180, 30, 255]),
        "label": "plastic waste",
        "confidence": 0.6
    }
}

def analyze_texture_for_pollution(img):
    """
    Analyze image texture to detect unnatural patterns that could be debris.
    Uses edge detection and morphological operations to find potential plastic debris.
    Enhanced for better plastic bottle detection.
    
    Args:
        img: OpenCV image in BGR format
        
    Returns:
        List of bounding boxes for potential debris based on texture
    """
    try:
        # Convert to grayscale
        gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
        
        # Apply Gaussian blur to reduce noise
        blurred = cv2.GaussianBlur(gray, (5, 5), 0)
        
        # Apply Canny edge detection
        edges = cv2.Canny(blurred, 50, 150)
        
        # Dilate edges to connect nearby edges
        kernel = np.ones((3, 3), np.uint8)
        dilated_edges = cv2.dilate(edges, kernel, iterations=2)
        
        # Find contours in the edge map
        contours, _ = cv2.findContours(dilated_edges, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
        
        # Also do a more specific search for bottle-shaped objects
        # Convert to HSV for color filtering
        hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
        
        # Create a combined mask for common bottle colors
        bottle_mask = np.zeros_like(gray)
        
        # Clear/translucent plastic
        clear_mask = cv2.inRange(hsv, np.array([0, 0, 140]), np.array([180, 60, 255]))
        bottle_mask = cv2.bitwise_or(bottle_mask, clear_mask)
        
        # Blue plastic
        blue_mask = cv2.inRange(hsv, np.array([90, 40, 100]), np.array([130, 255, 255]))
        bottle_mask = cv2.bitwise_or(bottle_mask, blue_mask)
        
        # Apply morphological operations to clean up the mask
        bottle_mask = cv2.morphologyEx(bottle_mask, cv2.MORPH_CLOSE, kernel)
        bottle_mask = cv2.morphologyEx(bottle_mask, cv2.MORPH_OPEN, kernel)
        
        # Find contours in the bottle mask
        bottle_contours, _ = cv2.findContours(bottle_mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
        
        # Filter contours by various criteria to find unnatural patterns
        debris_regions = []
        
        # Process regular edge contours
        for contour in contours:
            # Calculate area and perimeter
            area = cv2.contourArea(contour)
            perimeter = cv2.arcLength(contour, True)
            
            # Skip very small contours
            if area < 100:
                continue
                
            # Calculate shape metrics
            if perimeter > 0:
                circularity = 4 * np.pi * area / (perimeter * perimeter)
                
                # Unnatural objects tend to have specific circularity ranges
                # (not too circular, not too irregular)
                if 0.2 < circularity < 0.8:
                    x, y, w, h = cv2.boundingRect(contour)
                    
                    # Calculate aspect ratio
                    aspect_ratio = float(w) / h if h > 0 else 0
                    
                    # Most natural objects don't have extreme aspect ratios
                    if 0.2 < aspect_ratio < 5:
                        # Get ROI and check for texture uniformity
                        roi = gray[y:y+h, x:x+w]
                        if roi.size > 0:
                            # Calculate standard deviation of pixel values
                            std_dev = np.std(roi)
                            
                            # Man-made objects often have uniform textures
                            if std_dev < 40:
                                debris_regions.append({
                                    "bbox": [x, y, x+w, y+h],
                                    "confidence": 0.55,
                                    "class": "Potential Debris (Texture)"
                                })
        
        # Process bottle-specific contours with higher confidence
        for contour in bottle_contours:
            area = cv2.contourArea(contour)
            if area < 200:  # Higher threshold for bottles
                continue
                
            perimeter = cv2.arcLength(contour, True)
            if perimeter <= 0:
                continue
                
            # Get bounding rectangle
            x, y, w, h = cv2.boundingRect(contour)
            
            # Calculate aspect ratio - bottles typically have aspect ratio between 0.2 and 0.7
            aspect_ratio = float(w) / h if h > 0 else 0
            
            # Bottle detection criteria - bottles are usually taller than wide
            if 0.2 < aspect_ratio < 0.7 and h > 50:
                # This is likely to be a bottle based on shape
                bottle_confidence = 0.70
                
                # Get ROI for additional checks
                roi_hsv = hsv[y:y+h, x:x+w]
                if roi_hsv.size > 0:
                    # Check for uniformity in color which is common in bottles
                    h_std = np.std(roi_hsv[:,:,0])
                    s_std = np.std(roi_hsv[:,:,1])
                    
                    # Bottles often have uniform hue and saturation
                    if h_std < 30 and s_std < 60:
                        bottle_confidence = 0.85  # Higher confidence for uniform color
                
                debris_regions.append({
                    "bbox": [x, y, x+w, y+h],
                    "confidence": bottle_confidence,
                    "class": "Plastic Bottle"
                })
        
        return debris_regions
    except Exception as e:
        logger.error(f"Texture analysis failed: {str(e)}")
        return []

def fallback_detect_objects(image_path: str) -> Dict:
    """
    Perform a simple color-based detection when ML detection fails.
    Uses basic computer vision techniques to detect potential pollution.
    
    Args:
        image_path: Path to the image file
    
    Returns:
        Dict with detections in the same format as the main detection function
    """
    if not HAS_CV2:
        logger.warning("OpenCV not available for fallback detection")
        return {"detections": [], "detection_count": 0, "annotated_image_url": None}
    
    try:
        # Read the image
        img = cv2.imread(image_path)
        if img is None:
            logger.error(f"Failed to read image at {image_path} in fallback detection")
            return {"detections": [], "detection_count": 0, "annotated_image_url": None}
        
        # Convert to HSV for better color detection
        hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
        
        # Make a copy for annotation
        annotated = img.copy()
        
        # Initialize detections
        detections = []
        
        # First check if the image contains water
        has_water = detect_water_body(image_path)
        logger.info(f"Water detection result: {'water detected' if has_water else 'no significant water detected'}")
        
        # Run texture-based detection for potential debris
        texture_detections = analyze_texture_for_pollution(img)
        detections.extend(texture_detections)
        logger.info(f"Texture analysis found {len(texture_detections)} potential debris objects")
        
        # Detect potential pollution based on color profiles
        for pollution_type, thresholds in COLOR_THRESHOLDS.items():
            # Create mask using HSV thresholds
            mask = cv2.inRange(hsv, thresholds["lower"], thresholds["upper"])
            
            # Apply some morphological operations to clean up the mask
            kernel = np.ones((5, 5), np.uint8)
            mask = cv2.morphologyEx(mask, cv2.MORPH_OPEN, kernel)
            mask = cv2.morphologyEx(mask, cv2.MORPH_CLOSE, kernel)
            
            # Find contours
            contours, _ = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
            
            # Filter out small contours, but with a smaller threshold for plastic debris
            if "plastic" in pollution_type:
                # More sensitive threshold for plastic items (0.5% of image)
                min_area = img.shape[0] * img.shape[1] * 0.005
            else:
                # Standard threshold for other pollution (1% of image)
                min_area = img.shape[0] * img.shape[1] * 0.01
                
            filtered_contours = [cnt for cnt in contours if cv2.contourArea(cnt) > min_area]
            
            # Process filtered contours
            for contour in filtered_contours:
                # Get bounding box
                x, y, w, h = cv2.boundingRect(contour)
                
                # Add to detections
                detections.append({
                    "class": thresholds["label"],
                    "confidence": thresholds["confidence"],
                    "bbox": [x, y, x + w, y + h]
                })
                
                # Draw on the annotated image
                cv2.rectangle(annotated, (x, y), (x + w, y + h), (0, 255, 0), 2)
                
                # Add label
                label = f"{thresholds['label']}: {thresholds['confidence']:.2f}"
                cv2.putText(annotated, label, (x, y - 10), 
                           cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 2)
        
        # Save the annotated image
        annotated_image_path = f"{image_path}_fallback_annotated.jpg"
        cv2.imwrite(annotated_image_path, annotated)
        
        # Return the results - would normally upload image in real implementation
        return {
            "detections": detections,
            "detection_count": len(detections),
            "annotated_image_path": annotated_image_path,
            "method": "fallback_color_detection"
        }
        
    except Exception as e:
        logger.error(f"Fallback detection failed: {str(e)}")
        return {"detections": [], "detection_count": 0, "annotated_image_url": None}
        
def detect_water_body(image_path: str) -> bool:
    """
    Simple detection to check if an image contains a large water body.
    This helps validate if the image is related to marine environment.
    
    Args:
        image_path: Path to the image file
    
    Returns:
        True if a significant water body is detected
    """
    if not HAS_CV2:
        return True  # Assume yes if we can't check
    
    try:
        # Read image
        img = cv2.imread(image_path)
        if img is None:
            return False
            
        # Convert to HSV
        hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
        
        # Water detection thresholds (blue/green tones)
        lower_water = np.array([90, 50, 50])  # Blue/green hues
        upper_water = np.array([150, 255, 255])
        
        # Create mask
        mask = cv2.inRange(hsv, lower_water, upper_water)
        
        # Calculate percentage of water-like pixels
        water_percentage = np.sum(mask > 0) / (mask.shape[0] * mask.shape[1])
        
        # Return True if water covers at least 30% of image
        return water_percentage > 0.3
        
    except Exception as e:
        logger.error(f"Water detection failed: {str(e)}")
        return True  # Assume yes if detection fails