Development of an advanced deep learning and neural network method for automatic early detection of mastitis in dairy cattle : A thesis submitted in partial fulfilment of the requirements for the Degree of Doctor of Philosophy at Lincoln University

Abstract

Mastitis, a costly and prevalent disease in dairy cows, reduces milk yield, quality, and animal welfare, while increasing treatment costs. Early detection, especially in the subclinical stage, is crucial for controlling the disease and maintaining milk production. This study explores advanced methods for early mastitis detection using data from a robotic milking system. Initially, Self-Organising Map (SOM) was employed to capture the mastitis spectrum. Then, Fuzzy C-Means (FCM) clustering was applied to the SOM map, identifying five health states—healthy, early subclinical, subclinical, late subclinical and clinical—despite the absence of specific labels for the subclinical stages. Building on the insights gained from unsupervised learning with SOM and FCM, we then trained a Bidirectional Long Short-Term Memory (BiLSTM) network to forecast the cow health state for the following day using supervised learning. The BiLSTM model showed high efficacy in forecasting cow health states, with precision, recall, and F1-scores for each state ranging from 0.92 to 1.00. This approach advances mastitis detection by integrating SOM-FCM for spectrum capture and BiLSTM for temporal forecasting, improving early diagnosis and enabling targeted interventions, thus promoting precision dairy farming and better economic outcomes.