Research On Recognition Of Typical Behaviors Of Prenatal Dairy Goats Based On Deep Learning

With the development of dairy goat breeding industry,intensification and refinement,the intelligent realization of dairy goat behavior monitoring is particularly important.At present,the prenatal management and feeding of dairy goats mainly depends on the actual experience of the breeder,i.e.,with the help of their understanding of the dairy goats,artificially observing of the dairy goats before production to judge their health and growth status,which increases the breeding the labor intensity of the staff and the low work efficiency are not fast enough to respond to changes in the individual behavioral status of the dairy goats,nor can they adapt well to the needs of large-scale farm feeding.With the continuous development of information technology,related workers have been carried out on intelligent feeding of dairy goats,and the observation and monitoring of their behaviour by image processing technology have made some achievements,but the requirement of this technology on environment is relatively high.In this paper,the intelligent wearing device with an attitude sensor was used,and the behaviour data of dairy goats was denoised by wavelet thresholds.Kalman Filtering(KF)was adopted to optimize the Spatial-Temporal Long Short Term Memory network(STLSTM)to establish the prenatal behaviour classification model.The main contents of the paper include:(1)Data acquisition and posture analysis.In order to obtain the behaviour data of dairy goat during the production period,this paper carries on the analysis and pretreatment of the data by intelligent wearables with attitude sensor attached to the body.The wavelet thresholding method was used to denoise four kinds of movement behaviour data,such as lying,standing,walking and digging,and the characteristic parameters of variance,mean value and peak value were selected to characterize the behaviour of dairy goat.Meanwhile,the movement posture angle and movement behaviour of the milk goats were analyzed by the Kalman filter algorithm,which provided the data foundation for abnormal behaviour before giving birth.(2)Classification and recognition of prenatal behavior of dairy goats based on optimized spatio-temporal long short-term memory network.To acquire spatiotemporal characteristics of dairy goat behavior and remove noise.In this paper,a Kalman filtering algorithm is proposed to optimize the integration of STLSTM and to establish a classification model of prenatal behaviour of dairy goats.In order to evaluate the performance of the behaviour classification algorithm—KF-STLSTM,our preprocessed behaviour attitude datasets of dairy goats were experimentally tested and compared with the classification results of the models,such as the Long Short Term Memory Network(LSTM)model and Convolution Neural Network(CNN)model.The results show that the KF-STLSTM algorithm has an average accuracy of 98.33 %,and is superior to the other four methods in accuracy and recall rate.(3)Design and implementation of prenatal behaviour management system for dairy goats.According to the demand research,The content of this paper includes using the Java and Matlab to program,using My SQL as the system database,designing and developing the B/S structure’s pre-natal goat sheep management system.The system completes the display of cached data,sensor management,and design of user information management modules.The functions of collecting typical prenatal behavior information,displaying and storing data are realized.The stability and operability of the built system are verified by the test.