智慧农业导刊2025，Vol.5Issue(11)：18-21,25,5.DOI:10.20028/j.zhnydk.2025.11.005
基于改进YOLOv5s的马脸识别方法研究
摘要
Abstract
In order to achieve rapid identification of horse identity,this paper proposes a lightweight detection algorithm based on YOLOv5s,using a self-built dataset as the research object.Firstly,the backbone network of YOLOv5s is replaced with the lightweight neural network MobileNetv3.Secondly,NAM,ParNet,and Triplet attention modules are added to the C3 module of the head network.Finally,the nearest neighbor interpolation upsampling method is replaced with transposed convolution upsampling.The optimal model(YOLOv5s+v3+Triplet+ConvTranspose2d)achieves an average precision mean of 99.5%,an accuracy of 97.2%,a recall of 98.9%,and a model size of 10.9 MB.Compared to the basic YOLOv5s model,the improved model exhibits an increase in accuracy by 0.5%,an increase in recall by 0.2%,and a reduction in model size by 3.9 MB.The modified model maintains a high level of performance while significantly reducing the model size,providing a methodological reference for the digitization and intelligentization of livestock farming,and thus holds high application value.关键词
马脸识别/YOLOv5s/MobileNetv3/注意力模块/转置卷积上采样Key words
horse face recognition/YOLOv5s/MobileNetv3/attention module/transposed convolution upsampling分类
计算机与自动化引用本文复制引用
张立娟,唐开婷..基于改进YOLOv5s的马脸识别方法研究[J].智慧农业导刊,2025,5(11):18-21,25,5.基金项目
2024-2025年度伊犁州职业教育研究项目(YLGZ012) （YLGZ012）
