Research Of Dairy Goat Behavior Classification Based On Multi-sensor Data

Dairy goat behavior recognition is one of the key technologies to realize the intelligent and refined breeding of dairy goats.The recognition of the behavior of dairy goats can help farmers timely understand the behavioral changes of dairy goats.So as to improve the management method,reduce the breeding cost,and provide basic data support for improving milk production and milk quality of dairy goats.Aiming at the problems that there are large individual differences in research objects,sensor technology defects,and few categories of behavior recognition in the process of using sensors for behavior recognition,in this paper multi-sensor technology,data fusion technology and multi-sensor-based behavior recognition algorithm were used to realize the behavior classification of dairy goats.The main research contents of this paper are as follows:(1)Multi-sensor collection of dairy goat behavior and environmental data.Through the analysis of dairy goat daily life environment and behavior characteristics,environmental monitoring modules and behavior monitoring modules are designed,and the collection of behavior data and environmental data of milk goats was realized by combining Wi-Fi communication and data transmission to the upper computer and cloud server.At the same time,the stress response and data stability of dairy goats at 6 different installation positions were compared and analyzed through experiments,and the best acquisition position was the front of the back.In the process of data collection,multiple sensors and cameras were used to record the behavior of milk goats at the same time,and the behavior labeling of milk goats was completed by comparing timestamps,and 360,000 tagged data records were finally obtained.(2)Dairy goat behavior data preprocessing.By combining quaternion and the method of complementary filtering,the behavior data fusion of dairy goat is realized,and the value of the acceleration without the component of gravity and attitude angle are obtained.Then,wavelet filter is used to denoise the fused data set and enhance the quality of the data set.Finally,combining with the periodicity rule of the behavior change of the dairy goat,the behavior recognition of the dairy goat is completed by the method of Fourier series fitting periodic function.A total of 12,000 sets of behavior data samples were obtained,which constituted the dairy goat behavior data set,and provided a good data basis for the behavior classification of dairy goats.(3)Dairy goat behavior recognition.In view of the characteristics of fast moving speed and diversified behaviors of dairy goats,in this paper two different behavior recognition schemes based on GBDT(Gradient Boosting Decision Tree,GBDT)algorithm and attentionbased Bi-LSTM(Bi-directional Long Short-Term Memory,Bi-LSTM)network were studied.The feature of regression tree and integrated learning strategy in gradient decision tree are studied and the behavior classification model is established.Experiments show that the average recognition accuracy of the algorithm is 92.4%.The recognition accuracy rate of the model for simple behaviors such as Lying-down and rest and standing up is about 93.3%,but that for complex behaviors such as circle movement,lameness and jumping is poor.In order to obtain a higher recognition accuracy rate of dairy goat's behavior,the Bi-LSTM network was studied,and the behavior classification model was established based on the attention mechanism.The experiment showed that the average recognition accuracy rate of the algorithm was 94.8%.Compared with GBDT model,the recognition accuracy rate of this model is improved by 6 percentage points.Because the time dependence of simple behavior is weak,the recognition accuracy of the model is not improved.According to the comprehensive analysis,the deep learning-based algorithm has obvious advantages over traditional machine learning in the recognition of complex behaviors.The recognition effect of the Bi-LSTM algorithm based on the attention mechanism basically meets the requirements of the behavior classification of dairy goats,and the recognition rate of weak time-dependent behaviors needs to be improved.