Development of a Wireless Measurement System for Classifying Cow Behavior Using Accelerometer Data and Location Data

Abstract

Abstract. A wireless measurement system for assessing behavior patterns in dairy cows has been designed and constructed. The system mainly consisted of 12 leg-tags, 6 location sensors, and a laptop computer. Twelve lactating Holstein dairy cows were selected in the trial. The leg-tags used a 433 MHz radio channel for transmitting the acceleration data and location data at 1 Hz. The data were logged to a laptop computer in real time. An ensemble classification algorithm was proposed in our study. The algorithm can be divided into two stages: at the first stage, the Semi-Supervised Fuzzy C-Means (SS-FCM) algorithm was used to interpret the acceleration data and classify all classes of behavior. Classified behavior patterns included feeding, lying, standing, lying down, standing up, normal walking, and active walking. Accuracy, sensitivity, and precision were used as statistical parameters of classification performance. The SS-FCM algorithm achieved a reasonable identification level of feeding (80% accuracy, 53% sensitivity, 56% precision), lying (92%, 96%, 89%), standing (90%, 51%, 58%), lying down (99%, 82%, 91%), standing up (99%, 66%, 78%), normal walking (97%, 95%, 86%), and active walking (99%, 95%, 88%). At the second stage, a D-S evidence theory method fused the result of SS-FCM algorithm and the cow position to classify all the behaviors predicted as feeding or standing at the previous stage. The sensitivity and precision of the two behaviors increased by an average of 17% and 16.5%, respectively. Overall, we found that the wireless measurement system provided an accurate, remote measure for cow behavior over the trial period, and the ensemble algorithm could effectively recognize various behavior patterns in dairy cows. Keywords: Behavior classification, Back propagation, D-S evidence theory, Received signal strength indication, Three-dimensional accelerometer.