Infrared Ship Instance Segmentation Method Based on Self-supervised Learning for Waterborne Traffic Surveillance
摘要
This study proposes an infrared ship instance segmentation method based on self-supervised learning (FIIIS-SSL), aimed at improving the segmentation of ship instances in infrared maritime surveillance images. By leveraging unlabeled infrared data, the framework combines semantic segmentation with self-supervised iterative learning to extract ship targets accurately, even in complex environments such as ports. Experimental results demonstrate a 4.89% improvement in average precision (AP) compared to existing methods, offering significant advancements in maritime traffic safety monitoring.