Optical-Guided Neural Collapse for SAR Few-Shot Class Incremental Learning

2026-06-03

AI Summary

This work addresses the challenges of large intra-class variation, inter-class confusion, and catastrophic forgetting in few-shot class-incremental learning (FSCIL) for synthetic aperture radar (SAR) imagery, which arise from data scarcity and aspect-angle sensitivity. To mitigate these issues, the authors propose an optical-guided FSCIL framework for SAR that leverages, for the first time, an orthogonal feature subspace learned from optical automatic target recognition (ATR) data as a geometric prior. SAR features are projected into this subspace via principal angle constraints and jointly optimized with a fixed simplex equiangular tight frame (ETF) classifier to induce neural collapse, thereby enhancing feature discriminability. Experiments on a 24-class SAR-optical ATR benchmark demonstrate that the proposed method significantly outperforms existing approaches such as NCFSCIL, achieving a superior trade-off between performance retention and final accuracy, with neural collapse metrics confirming improved intra-class compactness and inter-class separability.

Abstract

Few-shot class-incremental learning (FSCIL) in synthetic aperture radar imagery presents unique challenges due to severe data scarcity and SAR-specific variability. In particular, strong azimuth sensitivity in SAR induces large intra-class variation and inter-class confusion, and FSCIL sequential updates further lead to catastrophic forgetting of previously learned classes. Inspired by neural collapse, we propose an optical-guided SAR FSCIL framework, which derives orthogonal feature subspaces from a data-rich optical ATR dataset and uses them as geometric priors to guide SAR feature learning. SAR features are projected onto these orthogonal subspaces via principal angle constraints, effectively transferring discriminative structure from the optical to the SAR domain. Specifically, our projection loss and the classifier loss optimized with a frozen simplex-ETF geometry jointly induce neural collapse by concentrating features around class means while maintaining large inter-class angles. We evaluate the approach on a benchmark comprising an optical ATR dataset and a SAR ATR dataset with 24 target classes, organized into a base training session and seven incremental sessions. Compared with recent FSCIL methods including NCFSCIL and so on, our method achieves the highest final accuracy and a favorable trade-off between final performance and performance degradation. Moreover, neural collapse metrics show improved intra-class compactness and inter-class separability, indicating that the learned features more closely approximate the ideal simplex-ETF geometry.

Problem

• SAR
• few-shot class-incremental learning
• catastrophic forgetting
• azimuth sensitivity
• data scarcity

Innovation

• Neural Collapse
• Optical-Guided Learning
• Few-Shot Class-Incremental Learning
• Synthetic Aperture Radar (SAR)
• Simplex ETF Geometry