SAR Automatic Target Recognition

GNN-based model-architecture codesign for SAR automatic target recognition

This project develops high-performance, low-latency systems for Synthetic Aperture Radar (SAR) Automatic Target Recognition (ATR) using graph neural networks on FPGA. The work spans model design, hardware acceleration, and human-in-the-loop approaches.

Key contributions:

  • Proposed GNN-based models for SAR ATR that achieve state-of-the-art accuracy
  • Designed FPGA-based accelerators with model-architecture codesign for low-latency inference
  • Developed human-in-the-loop frameworks using reinforcement learning to improve ATR performance
  • Accelerated GNN-based SAR ATR on HBM-enabled FPGAs for high-bandwidth processing

Related publications:

  • B. Zhang, R. Kannan, C. Busart, V. Prasanna, “Model-Architecture Codesign for High-Performance and Energy-Efficient SAR ATR on FPGA,” IEEE Transactions on Geoscience and Remote Sensing, 2025.
  • B. Zhang, R. Kannan, V. Prasanna, C. Busart, “Accurate, Low-latency, Efficient SAR Automatic Target Recognition on FPGA,” IEEE FPL, 2022.
  • B. Zhang, R. Kannan, V. Prasanna, C. Busart, “Accelerating GNN-Based SAR Automatic Target Recognition on HBM-Enabled FPGA,” IEEE HPEC, 2023.