Few-Shot Class-Incremental Learning For Efficient SAR Automatic Target Recognition

TL;DR

SAR-ATR must rapidly adapt to new classes with limited labels while avoiding catastrophic forgetting. The authors propose DILHyFS, a dual-branch framework combining ResNet-18 and GFNet with cross-attention fusion, a fixed projection to dimension $M$, and a prototype-based incremental-LDA classifier guided by a hybrid loss $L = L_F + c L_C$, where $L_F$ is focal loss and $L_C$ is center loss. The approach leverages local feature extraction and long-range frequency-domain dependencies to improve discrimination under data scarcity, achieving state-of-the-art results on the MSTAR FSCIL-SAR-ATR benchmarks across multiple settings. The work demonstrates a data-efficient, robust pathway for continual SAR-ATR that mitigates forgetting while maintaining high accuracy, with practical implications for real-world surveillance and reconnaissance tasks.

Abstract

Synthetic aperture radar automatic target recognition (SAR-ATR) systems have rapidly evolved to tackle incremental recognition challenges in operational settings. Data scarcity remains a major hurdle that conventional SAR-ATR techniques struggle to address. To cope with this challenge, we propose a few-shot class-incremental learning (FSCIL) framework based on a dual-branch architecture that focuses on local feature extraction and leverages the discrete Fourier transform and global filters to capture long-term spatial dependencies. This incorporates a lightweight cross-attention mechanism that fuses domain-specific features with global dependencies to ensure robust feature interaction, while maintaining computational efficiency by introducing minimal scale-shift parameters. The framework combines focal loss for class distinction under imbalance and center loss for compact intra-class distributions to enhance class separation boundaries. Experimental results on the MSTAR benchmark dataset demonstrate that the proposed framework consistently outperforms state-of-the-art methods in FSCIL SAR-ATR, attesting to its effectiveness in real-world scenarios.

Figures (3)

• Figure 1: Overview of the proposed DILHyFS framework for FSCIL SAR-ATR.
• Figure 2: Illustration of the proposed feature fusion cross-attention module.
• Figure 3: Comparison with state-of-the-art methods in the 2-way 5-shot scenario.