Classification of the cattle behaviors by using magnitude and variance of accelerometer signal

Abstract

Time periods of walking-grazing, standing and lyingof cattle’s life can be used to predict their health. However, famer cannot observe those in all the time. Therefore, this paper proposes a simple technique to classify the cattle behaviors by using the magnitude and the variance of accelerometer output signal. There are two steps of algorithm detection, the first step employed the magnitude of each axis for classifying the cattle behaviors into two groups: 1) walking-grazing and standing and 2) lying. After that, the second step used the variance of Y-axis to notify between walking-grazing and standing behaviors. The classification results were inform time periods of each behavior and tested with two cattle. The measured precise times of each behavior were compared with human observation. As a result, we found that the detection testing can identify the cattlebehaviors with a high success rates, the system has the errors as follows walking-grazing maximum errors 2% standing maximum errors 13% and lying maximum errors 7%.