基于改进YOLOv8算法的垂钓活动监测方法  

A fishing behavior detection method based on an improved YOLOv8 algorithm Author:

作　　者：冯孟雅 FENG Mengya(Science and Technology Information Center of Anhui Province Pi-Shi-Hang Irrigation District Management Bureau,Lu'an 237000,China)

机构地区：[1]安徽省淠史杭灌区管理总局科技信息中心,安徽六安237000

出　　处：《江淮水利科技》2024年第4期46-50,共5页Jianghuai Water Resources Science and Technology

摘　　要：为实现对灌区垂钓活动的智能化精准识别,提出一种基于改进YOLOv8算法的垂钓活动监测方法。该方法在YOLOv8的骨干网络中添加多尺度特征融合模块(Conv-M),用来学习来自不同卷积层(Conv)的特征,同时利用自学习的权重系数对特征进行加权融合,增强网络对垂钓活动的特征提取能力。通过训练网络得到垂钓活动监测模型,实现对视频图像数据中的垂钓活动进行检测和识别。该方法相较于YOLOv8方法的查准率提高了1.1%,查全率提高了1.4%,平均识别精度提高了0.9%。研究成果可提高灌区垂钓活动监管的智能化水平。In order to realize the intelligent and accurate identification of fishing behavior in irrigation district,an improved CM-YOLOv8 fishing behavior detection method was proposed.This method added a multi-scale feature fusion module(ConvM)to the backbone network of YOLOv8 to learn features from different Conv layers.At the same time,the self-learning weight coefficient was used to weight the features,so as to enhance the ability of the network to extract features from fishing behavior.The fishing behavior detection model was obtained by training the network,and then the fishing behavior in the video image data was detected and recognized.Experimental results on Fish-Data showed that compared with YOLOv8,the proposed method can increase precision by 1.1%,and recall by 1.4%,and mean average precision increased by 0.9%.The results could improve the intelligent level of supervision of fishing behavior in irrigation district.

关 键 词：YOLOv8算法 Conv-M 数字灌区 垂钓活动监管 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]