Subclinical mastitis is one of the most prevalent and costly diseases in dairy farming, being laborious to detect by traditional methods. This exploratory study proposes a computational system for the early identification of subclinical mastitis in cattle by integrating infrared thermography (IRT) and machine learning (ML) algorithms. From thermal images of udders from Girolando and Jersey cows, 226 shape, texture, and statistical features were extracted. Various ML classifiers were evaluated, with emphasis on SVM models with polynomial and RBF kernels, which achieved AUC values around 97%, demonstrating a high ability to discriminate between healthy and diseased animals. The Extra Trees model reached 100% sensitivity, demonstrating IRT as a promising screening tool. The results validate IRT combined with ML as a non-invasive, low-cost, and automatable approach for monitoring herd health.
Louzada et al. (Tue,) studied this question.