Azimuth-Aware Discriminative Representation Learning for Semi-Supervised Few-Shot SAR Vehicle Recognition

Among the current methods of synthetic aperture radar (SAR) automatic target recognition (ATR), unlabeled measured data and labeled simulated are widely used to elevate performance SAR ATR. In view this, setting semi-supervised few-shot vehicle is proposed use these two forms cope with problem that few available, which a pioneering work in this field. allusion sensitivity poses vehicles, especially situation only data, we design azimuth-aware discriminative representation (AADR) losses suppress intra-class variations samples huge azimuth-angle differences, while simultaneously enlarging inter-class differences same azimuth angle feature-embedding space via cosine similarity. Unlabeled from MSTAR dataset pseudo-labels categories among SARSIM SAMPLE dataset, taken into consideration loss. The experimental settings randomly selected training set. phase amplitude targets all article. method achieves 71.05%, 86.09%, 66.63% under 4-way 1-shot EOC1 (Extended Operating Condition), EOC2/C, EOC2/V, respectively, overcomes other learning (FSL) (SSFSL) classification accuracy.