Cattle Collar: An End-to-End Multi-Model Framework for Cattle Monitoring

Abstract

Individual cattle behaviour monitoring is a promising way of improving cattle farm management by detecting health issues and anomalies in behaviour patterns. Accelerometer sensors are non-invasive, low-cost devices that track daily activities and behaviour. For this, a hardware setup is attached to the neck collar of the cow to record its behaviour. We proposed an efficient data labelling method to classify simultaneously occurring activities with a single inertial sensor and a temperature sensor. Then a Machine Learning (ML) model is trained to predict the cattle activities based on different time and frequency domain-based statistical features. The proposed method shows an accuracy of 86% for Random Forest classifier. The behavioural analysis of an individual cow is sent to the user interface. The application provides visual data representation to monitor multiple cows daily and weekly.