app.py

import gradio as gr
from PIL import Image, ImageDraw, ImageFont
import numpy as np
import os

# Check if model file exists
model_path = "export.pkl"
learn_inf = None

if os.path.exists(model_path):
    try:
        from fastai.vision.all import load_learner
        # Load your trained model
        learn_inf = load_learner(model_path)
        print("✅ Model loaded successfully!")
    except Exception as e:
        print(f"❌ Error loading model: {e}")
        print("This might be due to fastai version compatibility issues.")
        learn_inf = None
else:
    print("Warning: export.pkl not found. Please upload your trained model file.")

# Classes
classes = [
    "AnnualCrop","Forest","HerbaceousVegetation","Highway","Industrial",
    "Pasture","PermanentCrop","Residential","River","SeaLake"
]

# Assign colors for visualization
class_colors = {
    "AnnualCrop": (255, 255, 0),
    "Forest": (34, 139, 34),
    "HerbaceousVegetation": (144, 238, 144),
    "Highway": (128, 128, 128),
    "Industrial": (255, 165, 0),
    "Pasture": (173, 255, 47),
    "PermanentCrop": (0, 255, 0),
    "Residential": (255, 0, 0),
    "River": (0, 191, 255),
    "SeaLake": (0, 0, 139)
}

# Debug: Print class information
print(f"Total classes defined: {len(classes)}")
print(f"Total colors defined: {len(class_colors)}")
print("Classes:", classes)
print("Colors:", list(class_colors.keys()))

patch_size = 128
stride = 128

def classify_image(img: Image.Image):
    if learn_inf is None:
        return None, "❌ Model not loaded! Please check the console for error messages. You may need to retrain your model with a compatible fastai version or use fastai<2.8.0."
    
    try:
        img_np = np.array(img.convert("RGB"))
        h, w, _ = img_np.shape

        mask = np.zeros((h, w, 3), dtype=np.uint8)
        coverage = {cls:0 for cls in classes}

        # Sliding window classification
        for y in range(0, h-patch_size+1, stride):
            for x in range(0, w-patch_size+1, stride):
                patch = img_np[y:y+patch_size, x:x+patch_size]
                patch_pil = Image.fromarray(patch).convert("RGB")
                pred, _, _ = learn_inf.predict(patch_pil)
                color = class_colors[pred]
                mask[y:y+patch_size, x:x+patch_size] = color
                coverage[pred] += patch_size*patch_size

        # Blend original + mask
        blended = (0.6*img_np + 0.4*mask).astype(np.uint8)
        blended_img = Image.fromarray(blended)

        # Compute coverage percentages
        total_pixels = h*w
        coverage_pct = {cls: (coverage[cls]/total_pixels)*100 for cls in classes}

        # Prepare statistics text
        stats_text = "\n".join([f"{cls}: {coverage_pct[cls]:.2f}%" for cls in classes])

        return blended_img, stats_text
    except Exception as e:
        return None, f"❌ Error during classification: {str(e)}"

def create_color_legend():
    """Create a color legend image showing all land cover classes and their colors"""
    # Create a white background image - increased height to fit all 10 classes
    legend_width = 450
    legend_height = 450  # Increased height to accommodate all classes
    legend_img = Image.new('RGB', (legend_width, legend_height), 'white')
    draw = ImageDraw.Draw(legend_img)
    
    # Try to use a default font, fallback to basic if not available
    try:
        font = ImageFont.truetype("/System/Library/Fonts/Arial.ttf", 14)  # Slightly smaller font
    except:
        try:
            font = ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf", 14)
        except:
            font = ImageFont.load_default()
    
    # Title
    draw.text((10, 10), "🎨 Color Legend - Land Cover Classes", fill='black', font=font)
    draw.text((10, 30), f"Total Classes: {len(class_colors)}", fill='blue', font=font)
    
    # Draw color squares with labels - adjusted spacing
    y_offset = 55
    for i, (cls, color) in enumerate(class_colors.items()):
        # Draw color square
        y_start = y_offset + i * 35  # Increased spacing between rows
        y_end = y_start + 25
        
        # Draw the colored rectangle
        draw.rectangle([20, y_start, 50, y_end], fill=color, outline='black', width=2)
        
        # Add class name with index for debugging
        class_text = f"{i+1}. {cls}"
        draw.text((60, y_start + 2), class_text, fill='black', font=font)
        
        # Add RGB values for reference
        rgb_text = f"RGB{color}"
        draw.text((280, y_start + 2), rgb_text, fill='gray', font=font)
        
        y_offset += 35  # Increased spacing
    
    # Add footer with total count
    footer_y = y_offset + 10
    draw.text((10, footer_y), f"Legend shows {len(class_colors)} land cover classes", fill='green', font=font)
    
    return legend_img

# Build Gradio interface
with gr.Blocks(title="🌍 Satellite Land Cover Classifier") as iface:
    gr.Markdown("# 🌍 Satellite Land Cover Classifier")
    gr.Markdown("Upload a satellite image and get land-cover classification heatmap + coverage stats.")
    
    with gr.Row():
        with gr.Column(scale=2):
            # Input section
            gr.Markdown("## 📤 Upload Image")
            input_image = gr.Image(type="pil", label="Upload Satellite Image")
            classify_btn = gr.Button("🔍 Classify Image", variant="primary")
        
        with gr.Column(scale=1):
            # Color legend section
            gr.Markdown("## 🎨 Color Legend")
            legend_image = gr.Image(value=create_color_legend(), label="Land Cover Classes", interactive=False)
    
    with gr.Row():
        with gr.Column(scale=2):
            # Output section
            gr.Markdown("## 📊 Results")
            output_image = gr.Image(type="pil", label="Classification Result")
            output_stats = gr.Textbox(label="Coverage Statistics", lines=10)
        
        with gr.Column(scale=1):
            # Additional info
            gr.Markdown("## ℹ️ How it works")
            gr.Markdown("""
            This classifier uses a sliding window approach:
            - **Patch Size**: 128x128 pixels
            - **Stride**: 128 pixels (no overlap)
            - **Output**: Color-coded heatmap + coverage percentages
            
            **Note**: You need to upload your trained model file (export.pkl) to use this classifier.
            """)
    
    # Connect the button
    classify_btn.click(
        fn=classify_image,
        inputs=input_image,
        outputs=[output_image, output_stats]
    )

iface.launch()