Event-Driven sEMG Feature Extraction for Classifying Ingestive Behavior in Ruminants

Precision livestock farming (PLF) has motivated the development of sensor technology to monitor animal health and productivity. Feed intake and rumination time are crucial parameters, with surface Electromyography (sEMG) showing promise for differentiating between eating and ruminating chews. This letter investigates if features extracted from fixed-window sEMG segments improve classification accuracy compared to variable-length windows based on chew onset and offset. Data were collected from two Jersey cows using sEMG electrodes placed on the masseter muscle. Eating and rumination activities were visually labeled. The sEMG signal was segmented using a double threshold onset segmentation (DTOS) method based on the onset of muscle activity. Features were extracted from both fixed and variable-length windows defined by the DTOS segmentation. A linear discriminant analysis classifier was used to differentiate between eating and rumination chews. Results showed that features extracted from fixed-window segments achieved a significantly higher classification accuracy (87.85%) compared to variable-length windows. This approach simplifies feature extraction and potentially improves real-time classification for ingestive behavior monitoring in PLF systems.