基于改进YOLOv8的渔港船舶进出港目标检测与统计方法  

Detection and statistics method of ship entry and exit in a fishing port based on improved YOLOv8

作　　者：惠卓凡 李鹏龙 沈烈 沈辉 隋江华 张胜茂[2] HUI Zhuofan;LI Penglong;SHEN Lie;SHEN Hui;SUI Jianghua;ZHANG Shengmao(College of Navigation and Naval Engineering,Dalian Ocean University,Dalian 116023,China;Key Laboratory of Fishery Remote Sensing,Ministry of Agriculture and Rural Affairs,East China Sea Fisheries Research Institute,Chinese Academy of Fishery Sciences,Shanghai 200090,China;Liaoning Provincial Marine and Fishery Administrative Law Enforcement Corps,Dalian 116023,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试验室,上海200090 [3]辽宁省海洋与渔业行政执法总队,辽宁大连116023

出　　处：《大连海洋大学学报》2024年第3期498-505,共8页Journal of Dalian Ocean University

基　　金：国家自然科学基金(61936014)。

摘　　要：为减轻渔港执法人员在渔船进出港统计工作中的压力,提出了一种基于改进YOLOv8模型的渔港船舶进出港目标检测与统计模型YOLOv8n-Bi-FPN-EMA,该模型采用加权双向特征金字塔网络结构(Bi-FPN),引入高效多尺度注意模块(EMA)以提高模型的目标检测准确性和稳定性,并结合BoT-SORT跟踪算法实现对不同类型船舶进出渔港的分类统计。结果表明:与YOLOv8n模型相比,YOLOv8n-Bi-FPN-EMA模型的准确率提高了5.0%,mAP@0.5提高了0.2%,对执法船、拖船和渔船的识别准确率分别提高了9.9%、4.9%、0.3%。其中,对渔船的识别准确率为99.1%,召回率为84.4%,mAP@0.5为97.8%。研究表明,YOLOv8n-Bi-FPN-EMA模型在渔船识别上有较好的表现,可用于渔船识别与进出港统计的辅助工作。In order to alleviate the working pressure of law enforcement officers in fishing ports in the statistics of fishing vessels entering and leaving the port,a target detection and statistical model YOLOv8n-Bi-FPN-EMA for ships entering and leaving in a fishing port was proposed based on the improved YOLOv8n model.In the model YOLOv8n-Bi-FPN-EMA,the weighted bidirectional feature pyramid network structure(Bi-FPN)was adopted to optimize the feature fusion method of the YOLOv8 network model,and the efficient multi-scale attention module(EMA)were adopted was integrated.The YOLOv8 network model was introduced to improve the accuracy and stability of target detection,and the statistics of the entry and exit of various types of ships in the fishing port were realized through BoT-SORT.Compared with the YOLOv8n model,the accuracy of the YOLOv8n-Bi-FPN-EMA model was increased by 5.0%,mAP@0.5 improved by 0.2%,the recognition accuracy for law enforcement vessels,tugboats,and fishing vessels increased by 9.9%,4.9%,and 0.3%,respectively.the YOLOv8n-Bi-FPN-EMA network model was featured by increase in the accuracy by 5%,mAP@0.5 by 0.2%,by 9.9%for the identification accuracy of law enforcement vessels,by 4.9%for tugboats,and 0.3%for fishing vessels,with identification accuracy of 99.1%for fishing vessels,the recall rate of 84.4%,and the mAP@0.5 of 97.8%,indicating With the identification accuracy for fishing vessels of 99.1%,a recall rate of 84.4%and an mAP@0.5 of 97.8%.Research indicates that the YOLOv8n-Bi-FPN-EMA model has good performance in fishing vessel identification,and can be used for auxiliary work of fishing vessel identification and port entry and exit statistics.

关 键 词：深度学习 目标检测 YOLOv8 渔船识别 进出港统计 伏渔季管理 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] S981.1[自动化与计算机技术—计算机科学与技术]