Detection of Chewing Strokes from Jaw Movement Signals in Dairy Cows Using a Nose-Mounted Accelerometer.

This study evaluated a non-invasive nose-mounted accelerometer for automated detection of chewing strokes in dairy cows. Data were collected from 15 Holstein Friesians and validated against manual video annotations. Chewing strokes were identified using a peak detection algorithm applied to smoothed acceleration data. Two algorithm versions were analyzed: a raw version and a cleaned version that excluded a five-second interval during regurgitation, where no physiological chewing occurs. The cleaned version showed higher agreement with the reference method (Intraclass Correlation Coefficient [ICC] = 0.91; 95% Confidence Interval [CI]: 0.77-0.96) and lower error metrics (Mean Absolute Error [MAE]: 3.67; Root Mean Square Error [RMSE]: 4.72; Mean Absolute Percentage Error [MAPE]: 5.64%) compared to the raw version (ICC = 0.67; MAE: 10.00; RMSE: 11.48; MAPE: 15.27%). Both methods demonstrated that reliable detection of chewing activity is feasible using this sensor system. Automated chewing stroke detection may contribute to the assessment of rumen function, feeding behaviour, and animal welfare and may support future precision livestock farming applications by providing objective information on chewing activity.