基于多尺度特征融合的牛只异常毛色识别算法  

Multi-scale feature fusion-based algorithm for identifying abnormal coat colour in cattle

作　　者：杜广振 房建东[1,2,3] 王秀玲 赵于东[2,3] DU Guang-zhen;FANG Jian-dong;WANG Xiu-ling;ZHAO Yu-dong(College of Information Engineering,Inner Mongolia University of Technology,Hohhot 010080,China;Key Laboratory of Perception Technology and Intelligent System,Inner Mongolia Autonomous Region,Hohhot 010080,China;Collaborative Innovation Center of Perception Technology in Intelligent Agriculture and Animal Husbandry,Inner Mongolia Autonomous Region,Hohhot 010080,China)

机构地区：[1]内蒙古工业大学信息工程学院,内蒙古呼和浩特010080 [2]内蒙古自治区感知技术与智能系统重点实验室,内蒙古呼和浩特010080 [3]内蒙古自治区智慧农牧业感知技术协同创新中心,内蒙古呼和浩特010080

出　　处：《计算机工程与设计》2025年第3期682-689,共8页Computer Engineering and Design

基　　金：内蒙古自治区科技攻关基金项目(2019GG334、2019FF337);内蒙古自治区直属高校基本科研业务费基金项目(JY20220012)。

摘　　要：针对通过毛色特征来判断牛只健康状态时,现有技术存在模型参数大和识别精度低的问题,提出一种基于多尺度特征融合的牛只异常毛色识别算法。采用改进YOLOv5模型的方法对牛只异常毛色进行分类识别,更换GhostNet骨干网络,使模型更加轻量化;添加Biformer注意力机制,更换Bi-FPN颈部网络,更换EIoU损失函数,提高模型对不同种类毛色识别的准确性。实验结果表明,改进后网络的平均准确度为94.6%,相比原始YOLOv5模型提高7.1%,参数量减少22.9%,模型更加轻量化。Aiming at the problems of large model parameters and low recognition accuracy in the existing technology when judging the health status of cattle through hair colour features,a cattle abnormal hair colour recognition algorithm based on multi-scale feature fusion was proposed.The method of improving the YOLOv5 model was used to classify and identify the abnormal hair colour of cattle,and the GhostNet backbone network was replaced to make the model lightweight.The Biformer attention mecha-nism was added,the Bi-FPN neck network was replaced,and the EIoU loss function was replaced to improve the accuracy of the model in identifying different kinds of hair colours.The test results show that the average accuracy of the improved network is 94.6%,which is 7.1% higher than that of the original YOLOv5 model,and the amount of parameters is reduced by 22.9%,making the model more lightweight.

关 键 词：多尺度 智慧养殖 特征融合 注意力机制 细粒度 轻量化 动物福利 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]