| Using machine vision technology to automatically detect the behavior of heifers is beneficial to guarantee the quality of dairy cow reserve force and promote the development of intelligent breeding model.In this thesis,by analyzing the current situation of heifers detection technology at home and abroad,based on the demand of intelligent management dairy farms,the application of computer vision technology and deep learning method in behavior recognition and analysis of dairy cows was studied.The main work of this thesis is as follows:(1)In view of the small number of research samples on the behavior of heifer in China,the authors chose to make their own data sets,and labeled the data into five categories:lying down,crowding the head,eating,drinking and standing.So as to prepare Reasonable and accurate training data sets,validation data sets and test data sets for the behavior identification of heifer in this thesis.The dataset covers all the daily behaviors of the calves,and the complexity of the selected image background is consistent with the actual situation.(2)Faster R-CNN,YOLOv3 and YOLOv4,which are popular in deep learning,were studied.Meanwhile,the accuracy rate,recall rate,missed detection rate and false detection rate were used to evaluate and compare the three algorithms.The results showed that the YOLOV4 algorithm was better than the previous two algorithms in the recognition accuracy,which can meet the requirements of target recognition in the actual scene.(3)Based on the identification results of heifer’ behavior by YOLOv4 algorithm,the authors programed to statistical the occurrence frequency of various behaviors during the day.Then Logistic regression analysis was used to construct model for heifer’ abnormal behavior evaluation.The model evaluated the health condition of heifer by analyzing the time of lying down,head pressing,eating,drinking and standing,and realized the early warning of abnormal behavior of heifer.(4)Based on Qt Designer software,an online detection system was designed with Python programming to realize real-time identification of the behavior of heifer.The behavior recognition model constructed in this thesis has high accuracy and fast speed,which is suitable for application in actual scenes and has a broad application prospect.The abnormal behavior evaluation model of heifer can quickly and timely find the abnormality and provide reference for disease prevention of heifer. |
