基于改进YOLOv8n的安格斯牛面部识别  

Facial recognition of Angus cattle based on the improved YOLOv8n

作　　者：胡立俊 李旭[1] 李国亮 HU Lijun;LI Xu;LI Guoliang(Ministry of Education Key Laboratory of Tarim Oasis Agriculture/College of Information Engineering,Tarim University,Aral 510642,China;Ministry of Agriculture and Rural Affairs Key Laboratory of Smart Farming for Agriculltural Animals/College of Information,Huazhong Agricultural University,Wuhan 430070,China)

机构地区：[1]塔里木绿洲农业教育部重点实验室/塔里木大学信息工程学院,阿拉尔510642 [2]农业农村部智慧农业技术重点实验室/华中农业大学信息学院,武汉430070

出　　处：《华中农业大学学报》2025年第2期39-48,共10页Journal of Huazhong Agricultural University

基　　金：华中农业大学农业农村部智慧农业技术重点实验室开放项目(KLSFTAA-KF004);绿洲生态农业兵团重点实验室开放项目(202002)。

摘　　要：为解决安格斯牛独特的黑色毛发导致其面部特征区分困难的问题,采用基于YOLOv8n的改进方法,实现圈养环境中的安格斯牛准确、非接触式的面部识别。首先构建了一个包含200头安格斯牛在不同生长阶段的11 000张面部图像的数据集;其次,引入创新的增强感受野特征融合模块,该模块增强了模型对关键特征的关注;再次,设计了新型轻量化检测头LPCDH,用于安格斯牛的面部特征识别;最后,采用组泰勒剪枝方法,通过估计神经元的重要性剪除不重要的神经元,从而减少计算成本和内存占用,提升模型的部署效率。试验结果显示,改进后的模型平均识别准确率达到92.6%。与常用的SSD、YOLOv5n、YOLOv8s、YOLOv8m、YOLOv9t、YOLOv10n、RT-Detr和Mamba-YOLO模型相比,准确率分别提高了11.5、3.8、1.8、1.9、5.1、3.9、3.7和2.4百分点。与原始YOLOv8n模型相比,所设计模型在4折交叉验证中的准确率平均提高了3.1百分点。结果表明,该模型在内存消耗和计算需求方面实现了轻量化,特别适合在移动端和实际应用中的实时识别,可显著提高安格斯牛面部识别的准确率和效率。An improved YOLOv8n method was used for facial recognition of Angus cattle in captive environments to solve the problem of difficulty in distinguishing facial features caused by Angus cattle's unique black fur and to achieve the accurate and non-contact recognition.A dataset containing 11 000 facial images of 200 Angus cattle at different stages of growth was constructed.Introducing an innovative and enhanced receptive field feature fusion module was introduced to enhance the model's focus on key features.A novel lightweight detection head(LPCDH) was designed for recognizing the facial feature of Angus cattle.The group Taylor pruning method was used to eliminate irrelevant neurons by estimating their importance,thereby reducing computational costs and memory usage,and improving the deployment efficiency of the model.The results showed that the improved model achieved an average recognition accuracy of 92.6%,which was 11.5,3.8,1.8,1.9,5.1,3.9,3.7,and 2.4 percentage higher that of commonly used models including SSDs YOLOv5n,YOLOv8s,YOLOv8m,YOLOv9t,YOLOv10n,RT-Detr,and Mamba-YOLO model,respectively.The designed model was improved by 3.1 percentage in 4-fold cross-validation compared with the original YOLOv8n model.It is indicated that the constructed model is optimized for lightweight memory consumption and computational requirements,making it particularly suitable for real-time recognition on mobile devices and in practical applications,significantly improving the accuracy and efficiency of recognizing the facial feature of Angus cattle.It will have immense potential in individual recognition in the livestock industry.

关 键 词：安格斯牛 YOLOv8n 剪枝 牛只识别 面部检测模型 

分 类 号：S858.23[农业科学—临床兽医学] TP391.41[农业科学—兽医学]