面向海洋渔业捕捞生产的深度学习方法应用研究进展  

Research progress in the application of deep learning methods for marine fishery production:a review

作　　者：张胜茂 孙永文[1,2] 樊伟 唐峰华[2] 崔雪森[2] 伍玉梅[2] ZHANG Shengmao;SUN Yongwen;FAN Wei;TANG Fenghua;CUI Xuesen;WU Yumei(College of Navigation and Ship Engineering,Dalian Ocean University,Dalian 116023,China;Key Laboratory of Fisheries Remote Sensing,Ministry of Agriculture and Rural Affairs,East China Sea Fisheries Research Institute,Chinese Academy of Fishery Sciences,Shanghai 200090,China)

机构地区：[1]大连海洋大学航海与船舶工程学院,辽宁大连116023 [2]中国水产科学研究院东海水产研究所农业农村部渔业遥感重点实验室,上海200090

出　　处：《大连海洋大学学报》2022年第4期683-695,共13页Journal of Dalian Ocean University

基　　金：国家自然科学基金(61936014);中国水产科学研究院基本科研业务费(2020TD82);浙江省海洋渔业资源可持续利用技术研究重点实验室开放课题(2020KF001);WWF/OPF蔚蓝星球基金项目(P04593)。

摘　　要：随着全球渔业资源不断衰退,各国渔业机构和区域渔业管理组织采用渔业观察员方式促进渔业可持续捕捞,但人类观察员方式成本高、覆盖率低,难以满足管理需要。近年来,深度学习新模型不断涌现和完善,其检测速度、精度均在不断增强,为其应用于海洋渔业捕捞生产监控提供了条件。本文从数据获取、数据预处理、算法设计、模型训练和精度评价等方面,总结了渔业捕捞生产监测模型的构建过程,以渔船与渔船行为、渔获物、渔场预报、船员和渔具为对象,综述了深度学习技术在海洋渔业捕捞中的应用,并提出利用迁移学习或强化学习等方法来拓展识别目标种类及增强检测模型、利用高精度的特征提取网络提高目标分类准确率、通过边缘计算技术解决电子监控数据实时解析及制定统一标准以规范电子监控在渔业管理中的应用等未来重点研究方向,以期为深度学习在海洋渔业捕捞生产中的推广应用提供科学参考。With the continuous decline of global fishery resources,fishery agencies and regional fisheries management organizations of various countries use the fishery observer method to promote sustainable fishing,but the human observer method is characterized by so high cost and low coverage as to difficult to meet the management needs.In recent years,detection speed and accuracy have been continuously enhanced due to continuous emergence and improvement of new deep learning algorithms,thus providing conditions for marine fishery fishing production monitoring.The process of building a fishery production monitoring model is introduced from the aspects of data acquisition,data preprocessing,algorithm design,model training,and model accuracy evaluation.The application of deep learning technology in marine fishery fishing is discussed,and methods such as transfer learning or reinforcement learning are proposed in terms of fishing boats and boat behavior,catches,fishery forecasts,crew members,and fishing gear to expand the identification of target species and enhance detection algorithm,and use high-accuracy feature extraction network to improve the accuracy of target classification,to solve the real-time analysis of electronic monitoring data through edge computing technology and to formulate standards and specifications for electronic monitoring fishery management applications and other key research directions in the future,which provides reference for the promotion of deep learning in marine fishery fishing production.

关 键 词：深度学习 目标检测 渔获物识别与量测 卷积神经网络 海洋渔业捕捞生产 

分 类 号：S951.2[农业科学—水产养殖]