An Automated Approach for Lameness Detection in Cattle Using Machine Learning and IMU Sensors

Lameness in farm animals affects both their welfare and productivity. Current detection methods rely on subjective visual inspections. This study aims to develop an objective lameness detection method using sensor data from cattle limbs. Inertial Measurement Units (IMU) with a 500 Hz sampling rate were attached to cattle in Eastern Macedonia and Thrace, Greece. The sensors recorded movement data, which was processed to handle missing values. Time- and frequency-domain features were extracted from the data using a 2 s sliding window with 50 % overlap. ReliefF feature selection identified key gyroscope and accelerometer features. These features trained a MRF model to classify lameness severity into healthy, mild, and severe categories, achieving 93% accuracy. This study highlights the potential of sensor-based detection as an early warning system, advocating for refined IMU units and machine learning analysis as an objective alternative to visual inspections in farming.