Few-Shot SAR Target Classification via Instance-Level Awareness Contrastive Learning

Few-shot learning is a challenging task in synthetic aperture radar (SAR) target classification, where extensive labeled data is often unavailable. This paper presents a novel approach, Instance-Level Awareness Contrastive Learning, designed to enhance model representation by incorporating instance-level information and contrastive learning objectives. The proposed approach employs contrastive-based self-supervision to minimize the distance between each training sample and its rotated variants, enhancing sample discrimination and incorporating the instance level information. Furthermore, the model introduces a high degree of variability between base and rotated images through complex random transformations, enriching the feature information and improving the model’s generalization capabilities. Experiments on the MSTAR dataset demonstrate that the proposed method outperforms state-of-the-art few-shot approaches.