Inshore Ship Detection Based on Multi-Information Fusion Network and Instance Segmentation
Tian Tian, Peng Gao, Zhihong Pan, Hang Li, Lizhe Wang · 2020
Inshore ship detection is a challenging task due to the complex background and object placement in remote sensing port images. To address this problem, we propose a detection method based on semantic and instance segmentation. First, a multi-information fusion network is designed to segment ship objects by taking edge and global information into account, which comprises a multi-task network, a global network and a fusion network. Then a simple but effective instance segmentation method based on line scanning and interval analysis is presented to separate multiple ships within one semantic region. Experimental results on an inshore ship data set collected from Google Earth validate the state-of-the-art performance of the proposed method.