Deep learning-based classification of wakefulness, sleep, and rumination states in dairy cows from polysomnography and RumiWatch data

Accurate sleep monitoring is crucial for dairy cattle welfare, but the gold standard method, Polysomnography (PSG), is highly labor-intensive. To address this, researchers developed DairySleepNet, a deep learning framework designed to automatically classify wakefulness, sleep, and rumination states. Evaluated on over 77 hours of data from seven cows, the model processes signals from both PSG and the wearable RumiWatch System (RWS). Using PSG data, DairySleepNet achieved an impressive 90.80% accuracy, significantly outperforming baseline models and proving the viability of automated sleep annotation. However, combining PSG and RWS data did not enhance performance, and relying solely on RWS data yielded unsatisfactory results. These findings validate the success of automated PSG scoring but emphasize that further methodological advancements are required before accessible, sensor-based on-farm sleep monitoring becomes practical. Read the full article on Journal of Dairy Science.