Segmentation and density statistics of mariculture cages from remote sensing images using mask R-CNN

作     者：Chuang Yu Zhuhua Hu Ruoqing Li Xin Xia Yaochi Zhao Xiang Fan Yong Bai 

作者机构：School of Information and Communication EngineeringSchool of Computer Science&Cyberspace SecurityHainan UniversityNo.58Renmin AvenueHaikou 570228PR China Shenyang Institute of AutomationChinese Academy of SciencesShenyang 110016PR China 

出 版 物：《Information Processing in Agriculture》 (农业信息处理（英文）)

年 卷 期：2022年第9卷第3期

页      面：417-430页

核心收录：

学科分类：082803[工学-农业生物环境与能源工程] 0907[农学-林学] 0908[农学-水产] 08[工学] 09[农学] 0710[理学-生物学] 090801[农学-水产养殖] 0830[工学-环境科学与工程（可授工学、理学、农学学位）] 0707[理学-海洋科学] 0905[农学-畜牧学] 0828[工学-农业工程] 0906[农学-兽医学] 0829[工学-林业工程] 0901[农学-作物学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：This research was supported by the National Natural Science Foundation of China(Grant No.61963012) the Hainan Provincial Natural Science Foundation of China(Grant No.620RC564,Grant No.619QN195).The authors would like to thank the referees for their constructive suggestions 

主　　题：Deep learning Mask R-CNN Image segmentation Remote sensing 

摘      要：The normal growth of fishes is closely relevant to the density of mariculture. It is of greatsignificance to accurately calculate the breeding area of specific sea area from satelliteremote sensing images. However, there are no reports about cage segmentation and den-sity detection based on remote sensing images so far. And the accurate segmentation ofcages faces challenges from very large high-resolution images. Firstly, a new public mari-culture cage data set is built. Secondly, the training set is augmented via sample variationsto improve the robustness of the model. Then, for cage segmentation and density statistics,a new methodology based on Mask R-CNN is proposed. Using dividing and stitching tech-nologies, the entire remote sensing test images of the cage can be accurately ***, using the trained model, the object detection features and segmentation character-istics can be obtained at the same time. Considering only the area within the target detec-tion frame, the proposed method can count the pixels in the segmented area, which canobtain accurate area and density while reducing time-consuming. Experimental resultsdemonstrate that, compared with traditional contour extraction method and U-Net basedscheme, the proposed scheme can significantly improve segmentation precision and mod-el’s robustness. The relative error of the actual area is only 1.3%.