CLISAR-Net: A Deformation-Robust ISAR Image Classification Network Using Contrastive Learning

The inherent unknown deformations of inverse synthetic aperture radar (ISAR) images, such as translation, scaling, and rotation, pose great challenges to space target classification. To achieve high-precision classification for ISAR a deformation-robust image network using contrastive learning (CL), i.e., CLISAR-Net, is proposed deformation Unlike traditional supervised methods, CLISAR-Net develops new unsupervised pretraining phase, which means that the method uses two-phase training strategy In combined with data augmentation, positive negative sample pairs are constructed unlabeled then encoder trained learn discriminative deep representations images by CL. fine-tuning based on obtained from pretraining, classifier fine-tuned small number labeled finally, realized. experimental analysis, achieves higher accuracy than methods scaled, rotated, deformations. It implies learned more robust features through CL, ensures performance subsequent