How tonzowonzo/sar_utils works
a. Primary Data Flow:
• Raw SAR Image → Preprocessing → Model Selection → Despeckling → Output Image
b. Training Flow:
• Data Generator → Synthetic Speckle Creation → Model Training → Model Storage
Key Technical Elements to Highlight
a. Model Architecture Section:
• Dilation-based CNN
• Hybrid approach combining Wang and Zhang methods
• Optional skip connections for residual learning
b. Processing Components:
• Image scaling operations
• Linear/Log transformation handling
• Backscatter coefficient processing
Specific Implementation Details
• Input image requirements (0-255 scale)
• Processing time indicators (30s for 6000x6000 image)
• GPU/CUDA acceleration support
• Model selection logic
Additional Elements
a. Configuration Parameters:
• Image scaling thresholds
• Backscatter limits (0.2 threshold)
• Resolution handling
b. Performance Metrics:
• Processing time
• Model comparison results
Integration Points
• Interface with external libraries
• GPU acceleration points
• Potential expansion points for future implementations
Key Considerations: