Study on the monitoring method of sheep motion state in house feeding

In order to solve the problems of time-consuming and low efficiency of manual monitoring of house-fed sheep’s movement behavior, triaxial acceleration sensors were used to collect typical movement behavior data（feeding, walking and ruminating behaviors） of house-fed sheep. After filtering, feature extraction and principal component analysis, the behavior data were classified by support vector machine（SVM） algorithm and BP neural network algorithm, respectively. The results showed that both methods could effectively classify behavior data, among which the recognition rate of feeding behavior by SVM algorithm was as high as 97.3%, and the average recognition rate of behavior was 87.6%. Based on BP neural network algorithm, the recognition rates of feeding, walking and ruminating behaviors were 94.0%, 93.8% and 96.3%, respectively, and the average recognition rate of behaviors was 94.6%. It indicated that the BP neural network algorithm monitor the movement state of sheep in the house more effectively compared with the SVM algorithm, more effective than the support vector machine algorithm in monitoring the movement state of house-fed sheep, and the average recognition rate was higher.