Limited-Data SAR ATR Method Based on Causal Feature Extraction

This chapter introduces the limited-data SAR ATR methods based on causal feature extraction. Two kinds of methods based on causal feature extraction are proposed based on the counterfactual causal intervention architecture. Initially, a limited-data SAR inter-class stable causal feature extraction method is proposed, which alternately extracts and integrates local and global features. By leveraging stable inter-class distance loss and automatic data augmentation, this method enhances the discriminative power of SAR target features under limited-data conditions. Furthermore, a limited-data SAR region-searching causal feature extraction method is developed. Through the use of probability inversion image mask generation and effective region optimization search criteria, this method identifies and focuses on high-discriminative regions in SAR images, achieving true causal effects in feature extraction under limited-data conditions. The proposed methods are validated on real-world datasets, demonstrating their effectiveness in enhancing feature discrimination and performance in limited-data scenarios.