Cattle Behavior Classification Based on Integration of Acceleration Sensor and Position Information

In order to monitor the behavior of cattle, we designed an intelligent cattle collar, and completed the experiments at the National Breeding Center of Beef Cattle in YangLing, Shaanxi. The collar is composed of a control chip nRF52832 based on Bluetooth 5.0 and MPU6050 sensor, which can realize the collection of cattle behavior (acceleration, angular velocity) and position information. The micro controller calculates the position information of the cattle, according to the Received Signal Strength Indicator (RSSI) broadcast by the Bluetooth beacon in the cowshed. We set the dataset with 3-axis acceleration, 3-axis angular velocity and three-dimensional position coordinates as dataset A, and the dataset without adding location information as dataset B. These two datasets were input into SVM, KNN and DeepConvLSTM classification models to identify the walking, standing, eating and lying behaviors of cattle. The accuracy of dataset B on SVM, KNN and DeepConvLSTM is 84.65%, 98.96% and 99.13% respectively, while the accuracy of dataset A is 85.23
^%
,99.01 % and 99.53% respectively. Compared with dataset B, it has increased by 0.58%, 0.05% and 0.40%. At the same time, the F1 scores of walking, standing, eating and lying behaviors in dataset A are higher than those in dataset B. Among them, the F1 score of standing behavior is 0.77
^%
, 0.15% and 0.68% higher than that of dataset B in SVM, KNN and DeepConvLSTM classifiers. The experimental results show that the accuracy of cattle behavior classification is improved by adding position information. The integration of acceleration sensor and position information can monitor the behavior of cattle more accurately.