Global in Local: A Convolutional Transformer for SAR ATR FSL

Convolutional neural networks (CNNs) have dominated the synthetic aperture radar (SAR) automatic target recognition (ATR) for years. However, under limited SAR images, width and depth of CNN-based models are limited, widening received field global features in images is hindered, which finally leads to low performance recognition. To address these challenges, we propose a Transformer (ConvT) ATR few-shot learning (FSL). The proposed method focuses on constructing hierarchical feature representation capturing dependencies local each layer, named local. A novel hybrid loss interpret few forms labels contrastive image pairs, construct abundant anchor-positive anchor-negative pairs one batch provide sufficient optimization ConvT overcome sample effect. An auto augmentation enhance enrich diversity amount training samples explore hidden avoid over-fitting FSL. Experiments conducted Moving Stationary Target Acquisition Recognition dataset (MSTAR) shown effectiveness our Different from existing FSL methods employing additional datasets, achieved pioneering without other training.