TCSPANet: Two-Staged Contrastive Learning and Sub-Patch Attention Based Network for PolSAR Image Classification

Polarimetric synthetic aperture radar (PolSAR) image classification has achieved great progress, but there still exist some obstacles. On the one hand, a large amount of PolSAR data is captured. Nevertheless, most them are not labeled with land cover categories, which cannot be fully utilized. other annotating images relies more on domain knowledge and manpower, makes pixel-level annotation harder. To alleviate above problems, by integrating contrastive learning transformer, we propose novel patch-level classification, i.e., two-staged sub-patch attention based network (TCSPANet). Firstly, (TCNet) designed for representation information without supervision, obtaining discrimination comparability actual covers. Then, resorting to construct encoder (SPAE) modelling context within patch samples. For training TCSPANet, two datasets built up unsupervised semi-supervised methods. When predicting, algorithm, classifying or splitting, put forward realise non-overlapping coarse-to-fine classification. The results multi-PolSAR trained model suggests that our proposed superior compared