Research On Recognition Of Feeding Behavior Of Dairy Cows Based On Three-axis Acceleration Sensor

Animal husbandry is one of the important industries for Chinese farmers and merchants.Since ancient times,China has been very conducive to the development of animal husbandry topography and ecological environment,a large number of pastures.But for a long time,the high cost of dairy farming has plagued many herdsmen.How to realize accurate feeding and reduce the cost has become one of the difficult problems in the current dairy farming process.The traditional feeding mode is too extensive,resulting in a large amount of feed waste.In order to avoid this phenomenon,the monitoring of feed intake and feeding behavior of cows is indispensable.The feed intake of cows is closely related to the feed intake behavior of cows.Cow foraging behavior is generally divided into rolling,chewing,and other behaviors.Accurate identification of the three kinds of foraging behaviors is the basis of accurate feeding.This study attempts to establish a classification model of cow foraging behavior based on the data of triaxial acceleration collected in the process of cow foraging,so as to realize the automatic classification of cow foraging behavior.The work of this study is as follows:(1)modeling data acquisition.The experiment was carried out in a dairy farm in zhaozhuang village,yanqing district,Beijing.Thirteen cows with good body condition,normal food intake and similar size were selected as the research objects.In this study,the nasolabial levator muscle of 5 cows,the right masseter muscle of 5 cows,and the right masseter muscle of 3 cows(simultaneously sampled)were tested.The data acquisition frequency was 2Hz,and 120 sets of data were collected per minute.Each cow collected 3,600 sets of data on average,and a total of57,600 sets of triaxial acceleration data of feeding behavior were obtained.(2)data calibration and noise reduction.Using the methods of video calibration and weighing calibration,the three-axis acceleration of feeding behavior of cows was marked as rolling,chewing and other three feeding behaviors.On the basis of data calibration,this study conducted median filter + moving average,continuous wavelet transform,stationary wavelet transform and three methods for noise reduction for 21000 groups of 7 cows.The filtering effect was evaluated from three dimensions: standard deviation(SD),root mean square error(RMSE)and SNR.In general,continuous wavelet transform is the best denoising method with the highest SNR,the lowest root-mean-square error and the lowest standard deviation.(3)establishment and evaluation of the prediction model of cow foraging classification.In this paper,three kinds of cow foraging behavior classification models are established,which are probability neural network(PNN)classification model,BP neural network classification model and support vector machine(SVM)classification model.According to the classification prediction effect of the model,the recognition rate of the PNN model for the parts of bovine nasolabial muscle and right masseter was 53.99% and 61.5% respectively.In the data samples collected from 3 bovine nasolabial muscles and right masseter muscles at the same time,the recognition rates of bovine nasolabial muscles and right masseter muscles were 57.7% and51.1% respectively.The recognition rate of feeding behavior of the bovine nasolabial muscle and right masseter was 94.20% and 82.11%.In the data samples collected from 3 bovine nasolabial levator muscles and right masseter muscles at the same time,the recognition rates of bovine nasolabial levator muscles and right masseter muscles were 86.41% and 78.28% respectively.In the SVM classification model,the recognition rates of the nasolabial muscle and the right masseter muscle on the roll,chewing and other three kinds of foraging behaviors of cows were87.83% and 76.83% respectively.In the data samples collected from 3 bovine nasolabial muscles and right masseter muscles at the same time,the recognition rates of bovine nasolabial muscles and right masseter muscles were 86.11% and 82.5% respectively.In summary,among the three kinds of foraging behavior classification models of cows,the recognition rate of BP neural network and support vector machine is high,and the classification effect of the data acquired from the nasolabial muscle is better than that of the right masseter.The results of this study provide a theoretical basis for data collection,analysis and modeling of cow foraging behavior.