opensar-insight/rfi-detection-model

📋 RFI Segmentation Model

This folder contains inference model for binary segmentation of radio-frequency interference (RFI) in range-compressed SAR data, including:

• An unrestricted "U-Net" model used as the main or teacher model
• A lightweight "UNetSmall" model for faster deployment
• Smaller student variants and scratch training for ablation studies

These models were trained based on opensar-insight/rfi-detection-dataset.

🧩 Detailed List of Models

The main models available are:

• Unrestricted teacher model: BS64_lr3e-5_focal0.6_rfi_unrestricted.pth
• Lightweight student model: focal_lr1e-4_bs4_kdfeature_rfi_lightweight.pth

Further variants are available as part of ablation studies:

• Lightweight model trained from scratch: ablation_scratch_rfi_lightweight.pth
• Lightweight model with 6 base channels: ablation_basechannel6_rfi_lightweight.pth
• Lightweight model with 8 base channels: ablation_basechannel8_rfi_lightweight.pth

🔄 Input and Output

The input is a 4-channel SAR tensor built from VV/VH complex range-compressed data:

[VV_I, VV_Q, VH_I, VH_Q]

The output is a single-channel binary mask for the corresponding focused L1 SLC.

💻 Codebase

The full codebase, which includes training and inference capabalities, as well as script for further dataset generation, will soon be made open-source.