Detection Of Dairy Cow Lameness Based On Computer Vision And Deep Learning

Dairy cows lameness will cause a negative impact on large-scale dairy farming.Early detection and prevention are one of the effective methods to reduce the elimination rate of dairy cows in large-scale dairy farming.In order to solve the problems of artificial lameness detection in a large-scale breeding condition,and the difficulty to find mild lameness and sudden moderate and severe lameness behavior,this study takes dairy cows raised in dairy farms of Shaanxi Yangling Keyuan Clone co.,ltd.as the research object,and carries out the research on the detection method of dairy cow lameness behavior based on computer vision and in-depth learning,so as to meet the needs of upgrading the modernization level of dairy farming.The main research contents and conclusions of this thesis are as follows:(1)Aiming at the problems of time-consuming and labor-consuming in the actual detection of traditional dairy cow lameness behavior detection methods,a dairy cow lameness detection method based on double normal distribution background statistical model is proposed in this study.In order to verify the effectiveness of the lameness detection algorithm proposed in this study,the lameness detection experiment was carried out on the video set of dairy cows using the indexes of total accuracy,lameness accuracy and normal accuracy.The results showed that the total accuracy of D-Accuracy was 93.75%,that of L-Accuracy was 90.00%,and that of N-Accuracy was 100.00%.Compared with the typical GMM algorithm,the proposed algorithm was more suitable for detecting dairy cow lameness.The results also showed that the robustness against the environment was stronger and that the false positive rate of the target dairy cows was reduced by 18.71%.This method can provide new reference for fast and high-precision detection of dairy cow lameness.(2)According to the behavioral differences between lame dairy cows and normal dairy cows in back,head and leg,a Filter Layer based Yolov3(Filter Layer Yolov3,FLYOLOv3)deep learning network is proposed,and the detection method of key parts of cows in complex scenes is studied.In order to verify the effectiveness of the algorithm,it is compared with Faster R-CNN algorithm and YOLOv3 algorithm by using indicators such as precision,recall,average frame rate,and average accuracy.Test results showed that the precision of FLYOLOv3 algorithm was 98.21%,the recall was 96.32%,the average frame rate was 21 f/s,and the average precision was 92.17%.The precision of Faster R-CNN algorithm was 95.64%,the recall was 94.17%,the average frame rate was 10 f/s,and m AP was 91.49%.While the precision of YOLOv3 algorithm was 93.13%,the recall was 76.34%,the average frame rate was 85 f/s,m AP was 86.89%.For the key parts of dairy cows,the FLYOLOv3 algorithm is effective and feasible.(3)According to the variation features of the head and neck of the lame dairy cows during the movement process,a detection method of dairy cows lameness based on the normal distribution background statistical model(NBSM),the local circulation center compensation tracking model(LCCCT)and the linear slope nearest neighbor classification(DSKNN)technology is proposed.On the original data set,SVM,Naive Bayes and KNN classification algorithms are respectively used to carry out classification and detection tests on dairy cows lameness.The test results show that the classification and detection accuracy of SVM and Naive Bayes are 82.78% and the detection accuracy of mild lameness are 83.33%,66.67% and 83.33% respectively.After cleaning the original data set,the accuracy detection of lameness classification of KNN classification algorithm was 94.44%,and the accuracy detection of SVM classification and Bayes classification algorithm were 91.11%,and 86.11%,respectively.The above results show that the lameness of dairy cows can be correctly detected by fitting the slope features of straight lines at the joints of head,neck and back.After data cleaning,KNN classification algorithm can obtain better detection results.(4)According to the change features of the back curvature of the lameness dairy cows in the moving process,a lameness detection method of the dairy cows combining machine vision technology and depth learning algorithm is proposed.By detecting the back region of dairy cows on video,the curvature data of the target cow's back fitting curve is extracted and put into a classifier based on recurrent neural network to detect the lameness of dairy cows.In order to verify the effectiveness of the algorithm,567 videos were used to train the network model in LSTM model,Bi LSTM model,Noise+LSTM model and the model proposed in this paper respectively,and 243 videos were used for verification and testing.According to the fitting curvature data of dairy cow's back obtained by the algorithm used in this study,it is found that the average classification accuracy of the model proposed in this study is 96.61% in the parallel experiment of the classification detection of dairy cow lameness,which is 8.04%,2.09% and 5.78% higher than LSTM,Bi LSTM and Noise+LSTM models respectively,indicating that lameness of dairy cows can be correctly detected through analysis of the curvature features of dairy cows' backs.(5)Using GUI compiling toolbox in Matlab R2019 b,a dairy cow lameness behavior detection system in unstructured environment was designed.The software system mainly includes dairy cows lameness based on double normal distribution background statistical model,the cow key part detection algorithm based on FLYOLOv3,a dairy cow lameness behavior detection algorithm based on the dairy cow head and neck slope,and the dairy cow lameness behavior detection algorithm based on a dairy cow back curvature.After the test verification,the software system can realize the functions of detecting the lameness behavior of dairy cows through the above detection method,and the system has the advantages of simple operation,intuitive and specific processing process and result display,etc.