Unknown SAR Target Identification Method Based on Feature Extraction Network and KLD–RPA Joint Discrimination

Recently, deep learning (DL) has been successfully applied in automatic target recognition (ATR) tasks of synthetic aperture radar (SAR) images. However, limited by the lack SAR image datasets and high cost labeling, these existing DL based approaches can only accurately recognize training dataset. Therefore, precision identification unknown targets practical applications is one important capabilities that SAR–ATR system should equip. To this end, we propose a novel method for with joint discrimination. First all, feature extraction network (FEN) trained on dataset used to extract features, then are roughly identified from known computing Kullback–Leibler divergence (KLD) vectors. For cannot be distinguished KLD, their vectors perform t-distributed stochastic neighbor embedding (t-SNE) dimensionality reduction processing calculate relative position angle (RPA). Finally, finely RPA. Experimental results conducted MSTAR demonstrate proposed achieve higher accuracy than methods while maintaining targets.