摘要
由于水域监管面积大,钓鱼行为出现时间随机,导致传统的人工监管难度大、时效低。为了解决水域监管中钓鱼行为监管难的问题,提出了一种基于高精度轻量化的钓鱼行为识别方法,该方法主要利用YOLOv5人体识别模型提取人体图像,再对图像进行扩展、剪裁、标准化处理,最后对图像数据集进行样本筛选分类,利用MobileNetV3-Small模型对图像数据集进行训练获得钓鱼图像分类模型。MobileNetV3-Small钓鱼图像分类模型最高F1分数97.32%,查准率为96.58%,查全率为98.07%。通过YOLOv5人体识别模型和MobileNetV3-Small钓鱼图像分类模型实现了对钓鱼行为的高精度识别,解决了钓鱼行为人工监管难的问题,提高了监管效率。
Due to the large area of water regulation and the random occurrence of fishing behavior,traditional manual regulation is difficult and inefficient.In order to solve the problem of difficult regulation of fishing behavior in water regulation,a high-precision lightweight fishing behavior recognition method has been proposed.This method mainly uses the YOLOv5 human body recognition model to extract human body images,then extend,cuts,and standardizes the images.Finally,sample screening and classification are performed on the image dataset,and the MobileNetV3-Small model is used to train the image dataset to obtain a fishing image classi-fication model.The highest F1 score of the MobileNetV3-Small fishing image classification model is 97.32%,with a precision rate of 96.58%and a recall rate of 98.07%.Through the YOLOv5 human body recognition model and the MobileNetV3-Small fishing image classification model,high-precision recognition of fishing behavior has been achieved,solving the problem of difficult manu-al supervision of fishing behavior and improving regulatory efficiency.
作者
廖佳庆
胡嘉俊
李帆
LIAO Jia-qing;HU Jia-jun;LI Fan(Hangzhou Dingchuan Information Technology Co.,Ltd.,Hangzhou 310000,Zhejiang;Zhejiang Institute of Hydraulics and Estuary(Zhejiang Institute of Marine Planning and Design),Hangzhou 310000,Zhejiang;ZhejiangWater Conservancy Disaster Prevention and Reduction Quality Inspection Station,Hangzhou 310000,Zhejiang)
出处
《电脑与电信》
2023年第10期35-40,共6页
Computer & Telecommunication
基金
浙江省水利厅科技计划项目,项目编号:RC2155。
关键词
图像识别
深度学习
水域监管
非法钓鱼
image recognition
deep learning
regulation of water areas
illegal fishing
