A Novel Pseudo-Label Generation Method for Semi-Superivised SAR Target Recognition Based on Deep Learning

Automatic target recognition (ATR) is a significant application scenario for Synthetic aperture radar (SAR) image interpretation. Deep-learning based SAR ATR approaches relies on large amounts of labeled data. However, labeled SAR images are scarce and expensive. In this paper, a novel semi-supervised learning (SSL) framework is proposed for SAR ATR, which effectively alleviates the need of labeled samples. This method develops an epoch- and uncertainty-aware pseudo-label selection (EUAPS) mechanism, which takes advantage of the underutilized consistency between training epochs, and introduces the uncertainty estimates. EUAPS can select pseudo-labels with high-confidence, allowing the network to perform well even when labels are extremely scarce. Extensive experimental results illustrate that the EUAPS achieves 97.82% accuracy when 10 labeled samples per class and 87.86% when only 5 labels per class respectively, outperforming the state-of-the-art SSL methods and demonstrating the superiority.