Research On Multi-objective Real-time Dairy Cow Behavior Recognitin Of Video Surveillance

In recent years,influenced by the development of artificial intelligence and big data technology,the research in the domain of intelligent pasture has also been greatly developed,among which,the behavior recognition of dairy cows plays an important role in the development of intelligent pasture.In order to realize the automatic monitoring of cow behavior without harming cow animal welfare,the contactless cow behavior recognition method based on video surveillance has become a research hotspot in recent years.Therefore,this paper realizes the real-time recognition of standing,walking and lying behaviors of multiple cows in video surveillance by improving and combining a variety of deep learning algorithms.The main research contents are summarized as follows:First,based on the Open Pose model which is used for human pose estimation,modifying and applying it to dairy cow pose estimation to build the T-Open Pose model by Transfer Learning.And the T-Open Pose model is improved to obtain the position information of all key body parts of cows in a video image.Then,the YOLOv4 target detection algorithm is improved to improve the detection accuracy,and it is used to obtain target boxes position information of all cows in the image.Finally,a data processing method is proposed that normalizes the key body parts position information of cows in different locations in the image according to their own detection box.A classification network composed of fully connected layers is used to correspond key parts information to their respective behaviors under different behaviors to realize the recognition of standing,walking and lying behaviors.The cow behavior recognition model is trained and tested on the self-made dataset.The result shows that the recognition accuracies of standing,walking and lying can reach 94.00%,95.96% and 98.02%,and the FPS reaches 12.66,which meet the requirements of real-time behavior recognition,and provides a feasible new method for automatic behavior monitoring of diary cows in video surveillance.