An Integrated Counterfactual Sample Generation and Filtering Approach for SAR Automatic Target Recognition with a Small Sample Set

Although automatic target recognition (ATR) models based on data-driven algorithms have achieved excellent performance in recent years, the synthetic aperture radar (SAR) ATR model often suffered from degradation when it encountered a small sample set. In this paper, an integrated counterfactual generation and filtering approach is proposed to alleviate negative influence of The method consists component component. First, utilizes overfitting characteristics generative adversarial networks (GANs), which ensures samples. Second, built by learning different functions. component, multiple SVMs trained SAR sets provide pseudo-labels other improve rate. Then, improves dynamically while continuously generates At same time, samples that are beneficial also filtered. Moreover, ablation experiments demonstrate effectiveness various components method. Experimental results Moving Stationary Target Acquisition Recognition (MSTAR) OpenSARship dataset show advantages approach. Even though size constructed training set was 14.5% original set, reached 91.27% with