Cattle behaviour classification using 3-axis collar sensor and multi-classifier pattern recognition
Ritaban Dutta, Daniel Smith, RP Rawnsley, Greg J. Bishop-Hurley, James L. Hills · 2014
In this paper supervised machine learning techniques based multi-classifier pattern recognition system was developed and applied to classify cattle behavioural patterns recorded using collar systems fitted to individual dairy cows to infer their feeding behaviors. Cattle tag sensory system, consist of a piezoelectric micro-electromechanical chip containing a 3-axis accelerometer and a 3-axis magneto-resistive sensor (HMC6343 - Honeywell, Plymouth, MN), data were collected at the Tasmanian Institute of Agriculture (TIA) Dairy Research Facility in Tasmania. A multi-classifier pattern recognition system was applied to classify five common cattle behaviour classes, namely, Grazing, Ruminating, Resting, Walking, and Scratching. Part of the recorded cattle tag data were labeled with the known behavioural patterns observed by the field experimental scientists. Pattern recognition system had a sensory data preprocessor to extract window based statistical features from the time series data, and a supervised multi-classifier system to learn the extracted features and generate a model to classify unknown data into one of the five behaviour classes.