内蒙古民族大学学报（自然科学版）2024，Vol.39Issue(3)：58-66,9.DOI:10.14045/j.cnki.15-1220.2024.03.011
基于深度学习的奶牛乳头检测方法研究
Research on Detection Method of Cow's Nipple Based on Deep Learning
摘要
Abstract
In order to improve the accuracy and speed of cow's nipple target detection,the convolutional block attention module(CBAM)and lightweight convolution module(ghost module)were introduced into the YOLOv5 algo-rithm,and an improved CG-YOLOv5 target recognition algorithm was proposed.Improved methods:firstly,CBAM attention mechanism was introduced to learn and extract features related to cow's nipple target images.Secondly,the Ghost module lightens the C3 backbone network to reduce parameters and computation.Finally,the EIoU(effi-cient iou)loss function is used to replace the CIoU(complete iou)method to improve the regression accuracy and convergence speed of the model.The test results show that the improved CG-YOLOv5 target recognition algorithm performs well on the cow's nipple dataset,with an average detection accuracy of 92%and an inspection frame rate of 33.6,which is 4%and 16%higher than the original YOLOv5 algorithm respectively.This algorithm is superior to the original YOLOv5 algorithm in both detection accuracy and speed,verifying the applicability of the proposed CG-YOLOv5 algorithm in real-time detection of cow's nipple and other scenarios.关键词
YOLOv5/奶牛乳头数据集/CBAM注意力机制/Ghost模块/EIoU损失函数Key words
YOLOv5/cow's nipple dataset/CBAM attention mechanism/Ghost module/EIoU loss function分类
农业科技引用本文复制引用
席横流,王磊,王成军,夏事成..基于深度学习的奶牛乳头检测方法研究[J].内蒙古民族大学学报（自然科学版）,2024,39(3):58-66,9.基金项目
安徽省自然科学基金项目(2208085MF169) （2208085MF169）
安徽省高校协同创新项目(GXXT-2019-018) （GXXT-2019-018）
