A Trimodel SAR Semisupervised Recognition Method Based on Attention-Augmented Convolutional Networks

Semi-supervised learning (SSL) in synthetic aperture radars (SAR) is one of the research hotspots field radar-image automatic target recognition (ATR). It can efficiently deal with challenging environments where there are insufficient labeled samples and large unlabeled SAR dataset. In recent years, consistency regularization methods semi-supervised have shown considerable improvement accuracy efficiency. Current approaches suffer from two main shortcomings: firstly, extracting all relevant information image difficult owing to inability conventional convolutional neural networks (CNN) capture global relational information,; secondly, standard teacher-student methodology causes confirmation biases due high coupling between teacher student models. This work adopts an innovative tri-model method based on attention-augmented address aforementioned obstacles. Specifically, we develop attention mechanism incorporating a novel positional embedding recurrent (RNN), integrate this network as feature extractor, improve network's ability extract images. Further, bias problem by introducing classmate model structure utilise impose weak constraint (WCC) designed weaken strong-coupling student. Comparative experiments Moving Stationary Target Acquisition Recognition (MSTAR) dataset show that our outperforms state-of-the-art terms accuracy, demonstrating its potential new benchmark approach for deep community.