CCNR: Cross-regional context and noise regularization for SAR image segmentation

• DOI: 10.1016/j.jag.2023.103363
• Publication Date: 2023-06-08T16:41:04Z
• Authors: Zitong Wu; Biao Hou; Xianpeng Guo; Bo Ren; Zhonghua Li; Shuang Wang; Licheng Jiao
• Keywords: Robustness; Segmentation-based object categorization

Semantic segmentation, a fundamental research direction in synthetic aperture radar (SAR) image interpretation, has significant application value for multiple sectors. However, noise, multi-style terrains, geometric distortion, and shadows make SAR segmentation challenging. Although existing deep learning algorithms tend to mine the semantic relationship between pixels within individual images, they disregard global context of training data different regions from images. Moreover, noise resistance networks is not strong enough. All these factors degrade performance algorithms. In this study, algorithm using cross-regional regularization (CCNR) images proposed. CCNR three heads output results, representation features each pixel, reconstructed The self-attention contrastive are adopted explore region-level relations pixel-level achieving aggregation with same class. Furthermore, improve robustness network, applied impose additional constraints on encoder results. results experiments conducted large scene prove efficacy CCNR. Compared other comparative algorithms, proposed achieves more optimal increased robustness.