Study On Ruminant Behavior Recognition Of Dairy Cows Based On Activity Data And Neural Network

Nowadays,machine learning,artificial intelligence and other technologies are gradually combined with animal husbandry,which promotes the rapid development of intelligent animal husbandry.In the field of intelligent animal husbandry,the research on ruminants such as dairy cows is particularly extensive.The length of ruminating time can reflect the health status of ruminants.Therefore,if the identification and detection of the ruminating behavior of dairy cows can be realized,the abnormal ruminating cows can be found in time,the sick cows can be treated in time,and the loss of pasture can be avoided.In this paper,in order to solve the problems of difficult data extraction,too much noise and poor model recognition results in audio data and video data,based on the activity data of dairy cows and the algorithm knowledge in the field of neural network,the purpose of this paper is to identify the ruminating behavior of dairy cows.Firstly,this paper designs and implements an activity data acquisition system,which includes an activity collector,a communication gateway and a cloud platform.Through the data acquisition system,the activity data of more than 500,000 activity data of 25 cows with a minimum time granularity of one minute are collected.Secondly,the peak-to-peak values are used to extract the features of the activity data,and the ruminant state labels are obtained by using unsupervised clustering algorithms such as K-means and BIRCH and dimension reduction algorithms such as PCA and t-SNE.After comparison,the K-means+PCA model with the best effect is selected to obtain the labels,and the multi-classification labels are modified to two-classification labels to complete the pre-processing of the activity data.Finally,this paper analyzes the temporal correlation of cow ruminating behavior,and designs a cow ruminating behavior recognition algorithm model based on Long Short-Term Memory(LSTM)network.The appropriate network structure and parameters are determined,and the Softmax layer is added to obtain the probability distribution of classification.And the model is tested.The test results show that the cow ruminant behavior recognition algorithm has different recognition results for different data set partition methods.The recognition accuracy of the split single cow data partition method is 94.30%,and the recognition accuracy of the method for randomly selected individual dairy cows is 96.27%.When using the same test set,the recognition accuracy of LSTM model(96.27%)is better than traditional RNN model(92.89%).The average daily ruminating time of dairy cows in the data period was identified by the ruminant recognition algorithm based on LSTM.And ruminating time was fed back to the farm staff as the basis for judging the health of dairy cows.It verified the feasibility of this design and provided new basis and ideas for further research in the related fields of intelligent animal husbandry.