Efficient Instance Segmentation Paradigm for Interpreting SAR and Optical Images

Instance segmentation in remote sensing images is challenging due to the object-level discrimination and pixel-level for objects. In applications, instance adopts instance-aware mask, rather than horizontal bounding box oriented object detection, or category-aware mask semantic segmentation, interpret objects with boundaries. Despite these distinct advantages, versatile methods are still be discovered images. this paper, an efficient paradigm (EISP) interpreting synthetic aperture radar (SAR) optical proposed. EISP mainly consists of Swin Transformer construct hierarchical features SAR images, context information flow (CIF) interweaving from branch branch, confluent loss function refining predicted masks. Experimental conclusions can drawn on PSeg-SSDD (Polygon Segmentation—SAR Ship Detection Dataset) NWPU VHR-10 dataset (optical dataset): (1) Swin-L, CIF, acts whole utility; (2) EISP* exceeds vanilla R-CNN 4.2% AP value 11.2% dataset; (3) The poorly segmented masks, false alarms, missing segmentations, aliasing masks avoided a great extent segmenting images; (4) achieves highest compared state-of-the-art methods.