Instance segmentation ship detection based on improved Yolov7 using complex background SAR images

Frontiers in Marine Science

Marine Science

Mengge Liu¹  Lili Zhan¹  Jianhua Wan²  Muhammad Yasir²  Shanwei Liu²  Qamar Ul Islam³  Md Sakaouth Hossain⁴  Syed Raza Mehdi⁵  Arife Tugsan Isiacik Colak⁶  Qian Yang⁷ 

[1] College of Geodesy and Geomatics, Shandong University of Science and Technology, Qingdao, China;College of Oceanography and Space Informatics, China University of Petroleum (East China), Qingdao, China;Department of Electrical and Computer Engineering, College of Engineering, Dhofar University, Salalah, Oman;Department of Geological Sciences, Jahangirnagar University, Dhaka, Bangladesh;Department of Marine Engineering, Ocean College, Zhejiang University, Zhoushan, Zhejiang, China;National University International Maritime College Oman, Sahar, Oman;People's Liberation Army (PLA) Troops No.63629, Beijing, China;

关键词: computer vision;    object detection;    instance segmentation;    HR-RS;    YOLOv7;    SSDD;    HRSID;    SAR Complex background images;   

DOI  :  10.3389/fmars.2023.1113669

 received in 2022-12-01, accepted in 2023-04-07,  发布年份 2023

来源: Frontiers

【 摘 要 】

It is significant for port ship scheduling and traffic management to be able to obtain more precise location and shape information from ship instance segmentation in SAR pictures. Instance segmentation is more challenging than object identification and semantic segmentation in high-resolution RS images. Predicting class labels and pixel-wise instance masks is the goal of this technique, which is used to locate instances in images. Despite this, there are now just a few methods available for instance segmentation in high-resolution RS data, where a remote-sensing image’s complex background makes the task more difficult. This research proposes a unique method for YOLOv7 to improve HR-RS image segmentation one-stage detection. First, we redesigned the structure of the one-stage fast detection network to adapt to the task of ship target segmentation and effectively improve the efficiency of instance segmentation. Secondly, we improve the backbone network structure by adding two feature optimization modules, so that the network can learn more features and have stronger robustness. In addition, we further modify the network feature fusion structure, improve the module acceptance domain to increase the prediction ability of multi-scale targets, and effectively reduce the amount of model calculation. Finally, we carried out extensive validation experiments on the sample segmentation datasets HRSID and SSDD. The experimental comparisons and analyses on the HRSID and SSDD datasets show that our model enhances the predicted instance mask accuracy, enhancing the instance segmentation efficiency of HR-RS images, and encouraging further enhancements in the projected instance mask accuracy. The suggested model is a more precise and efficient segmentation in HR-RS imaging as compared to existing approaches.

【 授权许可】

Unknown   
Copyright © 2023 Yasir, Zhan, Liu, Wan, Hossain, Isiacik Colak, Liu, Islam, Raza Mehdi and Yang