基于深度学习的西北太平洋柔鱼渔场预测  

Prediction on fishing ground of Ommastrephes bartramii in Northwest Pacific based on deep learning

作　　者：解明阳 柳彬 陈新军[1,4,5,6] XIE Mingyang;LIU Bin;CHEN Xinjun(College of Marine Sciences,Shanghai Ocean University,Shanghai 201306,China;Key Laboratory of Marine Ecological Monitoring and Restoration Technologies,Ministry of Natural Resources,Shanghai 200137,China;State Key Laboratory of Satellite Ocean Environment Dynamics,Second Institute of Oceanography,Ministry of Natural Resources,Hangzhou 310012,China;Key Laboratory of Oceanic Fisheries Exploration,Ministry of Agriculture and Rural Affairs,Shanghai Ocean University,Shanghai 201306,China;National Engineering Research Center for Oceanic Fisheries,Shanghai Ocean University,Shanghai 201306,China;Key Laboratory of Sustainable Exploitation of Oceanic Fisheries Resources,Ministry of Education,Shanghai Ocean University,Shanghai 201306,China)

机构地区：[1]上海海洋大学海洋科学学院,上海201306 [2]自然资源部海洋生态监测与修复技术重点实验室,上海200137 [3]自然资源部第二海洋研究所,卫星海洋环境动力学国家重点实验室,浙江杭州310012 [4]上海海洋大学,农业农村部大洋渔业可持续利用重点实验室,上海201306 [5]上海海洋大学,国家远洋渔业工程技术研究中心,上海201306 [6]上海海洋大学,大洋渔业资源可持续开发省部共建教育部重点实验室,上海201306

出　　处：《水产学报》2024年第11期117-128,共12页Journal of Fisheries of China

基　　金：国家自然科学基金(NSFC42476086,NSFC42006159);上海市科技创新行动计划(19DZ1207502);卫星海洋环境动力学国家重点实验室资助项目(QNHX2238)。

摘　　要：柔鱼为西北太平洋主要的经济头足类物种,准确判别渔场空间分布能够更科学、有效地为渔业生产提供依据。在海洋渔业和海洋遥感的大数据时代下,如何提高模型精度、稳定性、计算效率,提取并挖掘有价值信息成为渔场学研究的挑战性问题。为此,本研究基于深度学习和渔场学理论,采用人工智能最前沿的U-Net模型,以海表面温度(SST)为输入因子,中心渔场分布为输出因子,构建了1998—2019年每年7—11月中心渔场预测模型。结果显示,验证集准确率为86.7%,训练集准确率为89.7%,测试集(2020年)的准确率、精确率、召回率和F1分数分别为87.2%、0.91、0.87和0.89,实际作业的渔获量数据与预测的中心渔场范围基本匹配,模型应用效果良好。在不同气候事件下,模型性能具有较好的适应能力。模型分析显示,厄尔尼诺事件下中心渔场纬度向南偏移,拉尼娜事件则相反。研究表明,本研究构建的U-Net模型可有效解决复杂数据下渔场预测的问题,提高渔场预测模型精度。本研究可为人工智能技术实现中心渔场的预报提供基础,具有很好的应用前景。Neon flying squid(Ommastrephes bartramii)is a primary economic cephalopod species in the Northwest Pacific Ocean.Accurate identification the spatial distribution of fishing ground provides a scientifically sound and effective foundation for fishery production.In the era of big data in marine fisheries and marine remote sensing,extracting and mining valuable information from vast datasets has emerged as a significant challenge in forecasting fishing grounds.Consequently,this study utilizes the theories of deep learning and fisheries oceanography,utilizing sea surface temperature(SST)data as input to develop a U-Net model for discriminating central fishing grounds from July to November in 1998-2019.The results indicate an accuracy of 86.7%for the validation set,89.7%for the training set,and the accuracy,precision,recall and balanced F1-score values for the 2020 test set being 87.2%,0.91,0.87 and 0.89,respectively.The catch data from fisheries is largely consistent with the predicted central fishing grounds,and the model's application proved effective.Across various climatic conditions,the model demonstrates robust adaptability.The latitude of the central fishery shifts southward during El Niño event and shifts northward during La Niña events.The model constructed in this study can effectively address the problem of fishery discrimination under complex data set,improve the precision of fishing ground prediction models,and lay a theoretical basis and foundation for the realization of fishing ground prediction.It holds promising application prospects.

关 键 词：柔鱼 渔场 深度学习 U-Net模型 西北太平洋 

分 类 号：S934[农业科学—渔业资源] TP181[农业科学—水产科学]