HTC+ for SAR Ship Instance Segmentation

Existing instance segmentation models mostly pay less attention to the targeted characteristics of ships in synthetic aperture radar (SAR) images, which hinders further accuracy improvements, leading poor performance more complex SAR image scenes. To solve this problem, we propose a hybrid task cascade plus (HTC+) for better ship segmentation. Aiming at specific task, seven techniques are proposed ensure excellent HTC+ scenes, i.e., multi-resolution feature extraction network (MRFEN), an enhanced pyramid net-work (EFPN), semantic-guided anchor adaptive learning (SGAALN), context ROI extractor (CROIE), mask interaction (EMIN), post-processing technique (PPT), and hard sample mining training strategy (HSMTS). Results show that each them offers observable gain, scenes becomes better. On two public datasets SSDD HRSID, surpasses other nine competitive models. It achieves 6.7% higher box AP 5.0% than HTC on SSDD. These 4.9% 3.9% HRSID.