Wearable IMU-based Early Limb Lameness Detection for Horses using Multi-Layer Classifiers

Objective, automated early lameness detection plays an important role for animal well-being. The work in this paper uses horse locomotion data collected by wearable inertial measurement units, extracts gait cycle routines and constructs a multi-layer classifier for horse lameness detection, identification, and evaluation. Multi-layer classifier (MLC) is based on support vector machine and K-Nearest-Neighbors methods. Each layer is independently designed and works as a binary classifier. Horse gait classification and limb lameness detection and evaluation are then handled by each layer successively. Experiment results show that the MLC achieves 94 % detection accuracy and also generates superior performance than a deep convolutional neural network-based multiclass classifier in terms of various assessment criteria.