Rumination Recognition Method Of Dairy Cows Based On The Noseband Pressure

Rumination is an important behavior of ruminants to convert plant resources into animal resources,and reflects the physiological health and welfare level of ruminants.Disease,calving,estrus,forage particle,stress response and so on can affect the duration of rumination,so we can monitor the changes of rumination to feedback the physiological health and welfare level of animals.Traditional artificial monitoring methods are time-consuming,laborious and inefficient.With the enlargement of breeding scale and the deepening of research o n precision animal husbandry,the demand for automatic rumination monitoring system is increasing.In this thesis,in order to realize the automatic monitoring of rumination,a set of noseband pressure signal acquisition equipment for dairy cows was design ed with pressure sensor as the core device.By analyzing the characteristics of noseband pressure signal,the short-term energy,standard deviation,shape index,root mean square envelope and extreme envelope were selected as the time-domain characteristics of noseband pressure signal,the frequency domain characteristics of noseband pressure signal were selected as the frequency spectrum of periodogram method,multi-window method and maximum entropy method,and the eigenmode function of noseband pressure signal was obtained by empirical mode decomposition.Kruskal-Wallis test method was used to calculate the distinction between different kinds of feature parameters.Without affecting the accuracy of model recognition,nine kinds of feature parameters were o ptimized.BP neural network model was used to verify the parameters that contribute little to the recognition accuracy,and the corresponding feature parameters were eliminated.The validation results showed that the parameters with small discrimination can be removed without affecting the recognition results,which effectively removed the redundant parameters.After optimizing the characteristic parameters,three kinds of pattern recognition methods,BP neural network,extreme learning machine and decision tree classifier,were used to recognize the rumination noseband pressure signal.The optimized combination of time-domain features,frequency-domain features,eigen-mode functions and feature parameters were used as input feature parameters,and then the recognition effect of the classifiers were analyzed and the optimal classifier was selected.The results showed that the recognition accuracy of decision tree classifier was higher than that of BP neural network and extreme learning machine.Therefore,thi s paper used decision tree classifier to identify the rumination noseband pressure signal in dairy cows.On the basis of decision tree as classifier,the recognition model of main parameters of rumination(number of rumination,duration of rumination and number of cuds)was constructed.The results showed that the model can identify the main rumination parameters very well.Compared with the rumination information observed directly,the recognition accuracy of the number of rumination,duration of rumination and number of cuds was 95.38%,94.67% and 92.87% respectively,and the average absolute percentage error was 1.41%,1.87% and 3.038% respectively,which showed that the method could effectively identify rumination.Based on decision tree classifier,this thesis realized the recognition of the number of rumination,duration of rumination and number of cuds of dairy cows of dairy cows,which provides a new method for automatic monitoring of rumination of dairy cows.