Semi-Supervised SAR ATR Framework with Transductive Auxiliary Segmentation

Convolutional neural networks (CNNs) have achieved high performance in synthetic aperture radar (SAR) automatic target recognition (ATR). However, the of CNNs depends heavily on a large amount training data. The insufficiency labeled SAR images limits and even invalidates some ATR methods. Furthermore, under few data, many existing are ineffective. To address these challenges, we propose Semi-supervised Framework with transductive Auxiliary Segmentation (SFAS). proposed framework focuses exploiting generalization available unlabeled samples an auxiliary loss serving as regularizer. Through segmentation information residue (IRL) training, can employ loop process gradually exploit compilation to construct helpful inductive bias achieve performance. Experiments conducted MSTAR dataset shown effectiveness our SFAS for few-shot learning. 94.18% be 20 each class simultaneous accurate results. Facing variances EOCs, ratios higher than 88.00% when 10 class.