Bi-Similarity Prototypical Network with Capsule-Based Embedding for Few-Shot SAR Target Recognition

This paper proposes a Bi -similarity prototypical network with capsule-based embedding to solve the problem of few-shot SAR target recognition. The proposed method comprises two procedures, i.e., feature embedding module and Bi-similarity reasoning module. Specifically, we build a feature embed-ding network with capsule operation, which can enable a feature embedding network to extract more informative features by effectively encoding relative spatial relationships between features. To reason the identity of target robustly, we develop a reasoning module based on Bi-similarity metric. Moreover, a mixed loss is proposed to train a discriminative representation space with both intra-class aggregation and inter-class separation. Experimental results on moving and stationary target acquisition and recognition (MSTAR) dataset show that the proposed method is effective and superior to some state-of-art methods in few-shot SAR target recognition tasks.

1. Introduction

Synthetic aperture radar (SAR) is a microwave sensor, which has been widely applied in civilian and military fields due to its operating characteristics of all-day, all-weather, and so on. Automatic target recognition (ATR) is one of the most important applications in SAR community [1]. In late years, deep learning techniques have attracted much attention in the field of SAR ATR. These are a great number of deep learning-ATR methods have been proposed. To name just a few, Wilmanski et al. [2] used modern training approaches adopted from the image processing literature for SAR ATR. Chen et al. proposed an all-convolutional networks (A-ConvNets) to learn features from SAR images automatically [3].