An Automatic Detection Method for Dairy Cow Teat Parameter Based on Improved PointNet++
SUN Junrui
ZHAO Kaixuan
WANG Jinjin
GAO Song
NIAN Yue
JIANG Shijie
Abstract：Dairy cow teat parameters are an important part of linear scoring.However,due to the small teat target and complex morphology of teats,achieving the rapid and accurate detection of multiple teat parameters remains challenging.This paper proposes an automatic detection method of dairy teat linear scoring parameters based on the improved PointNet++network model.Firstly,the Kinect DK camera was used to obtain the RGB-D data of the dairy cow scene,and the three-dimensional reconstruction was realized based on the ORB-SLAM2 algorithm fused with depth information.Secondly,in view of the characteristics of small teat target and difficult segmentation of dairy cows,the Heatblock attention mechanism was introduced to improve the segmentation accuracy of PointNet++network model.Finally,point cloud processing and principal component analysis were used to calculate the parameters of teat length,height difference between front and rear udders,and distance between front and rear teats.The results show that the segmentation accuracy of the improved model reaches 95.7％,with an average intersection over union of 80.2％,and the error of the parameters of the cow's teat is less than 3.85％.The method has the characteristics of high degree of automation,high precision and low stress on dairy cows,which can meet the requirements of automatic detection of dairy cow teat parameters.
Keywords：cow teat parameterPointNet++three-dimensional reconstruction
Publication Date：2025-10-25
Online Publishing Date：2025-11-17(First online date of this platform, not the publication date of the document)
Pages：10( 86-95 )