Cattle Behavior Patterns Classification Research Based On Improved RB-LSTM Network

The breeding of cattle is the key link of large-scale breeding industry,and the behavior of cattle is an important indicator to reflect the body condition and health of cattle.Therefore,cattle behavior monitoring is an important technology and tools in modern breeding.Deep learning-based research on livestock behavior recognition has grown in popularity in recent years.Intelligent wearable devices and deep learning algorithms can provide viable solutions for accurate and automatic monitoring of cattle behavior.In terms of intelligent livestock monitoring of diseases,advances in sensor technology and deep learning have the potential for helping in the early detection of infected cattle.Using deep learning and intelligent wearable devices to conduct research on automated collection of cattle behavior information,intelligent monitoring of cattle key behaviors,and disease warning,this paper focused on the problems of healthy breeding management of cattle as well as the development of information and intelligent monitoring.The following are the primary research results and findings:Firstly,to research the classification model of cattle behavior patterns including behavior patterns related to cattle dermatomycosis,activity data were collected from 12 cattle using a homemade collar equipped with an Inertial Measurement Unit(IMU).And a cattle behavior classification model based on the Long Short Term Memory-Recurrent Neural Network(LSTM-RNN)algorithm was established.The results show that the cattle behavior recognition and classification method based on IMU and time series data classification is feasible.Secondly,to improve the accuracy of cattle behavior recognition and classification,an improved Residual Bidirectional LSTM(RB-LSTM)model was built to recognize and classify cattle behavior patterns based on the successfully building of LSTM-RNN classification model in our previous experiments.Six cattle behavior patterns,namely,feeding,lying,ruminating(lying),rub itching(leg),social licking,and rub itching(neck)had classification accuracy of 98.1%,95.7%,85.4%,98.3%,96.6%,and 95.4%,respectively.Because two-way information can be used,the results show that the RB-LSTM network outperforms the baseline LSTM network in classifying time series data such as cattle behavior.Finally,aiming at the problems of real-time automatic monitoring equipment and practical farm management system in the development of large-scale breeding industry,based on the established classification model of cattle behavior pattern,combined with wireless transmission,Internet of Things and terminal technology,a smart pasture management system was designed to automatically identify key behaviors of cattle and count the time of various behaviors of individual cattle.And realize the early warning function of cattle skin mycosis.This paper realized the digital expression of physiological health variables of cattle based on deep learning and IMU equipment,which provides a feasible solution for the realization of large-scale cattle breeding and fine management.