Automatic recognition and classification of cattle chewing activity by an acoustic monitoring method with a single-axis acceleration sensor

Monitoring the eating habits and rumination of cattle is effective for evaluating forage values and making animal management decisions for stalls and grazing pastures. Acoustic monitoring is a feasible method to monitor these activities, but it requires automatic distinction and quantification of activities to gain broader application. This automatic distinction and quantification may be possible using a pattern matching method, which distinguishes jaw movements by extracting characteristic patterns from eating and ruminating, and then matches similar characteristic patterns in unanalyzed activities. The objectives of this study were to define an acoustic monitoring system with a single-axis acceleration sensor, to determine an effective sensor attachment site (i.e., the horn, forehead, or nasal bridge), and to assess the automatic classification of ingestive and ruminative chewing behaviors using a pattern matching method. Four beef cows fed Italian ryegrass silage were each recorded eating and ruminating for more than 60 and 20min, respectively. The oscillatory waveforms generated while eating and ruminating were clearly recorded by the single-axis acceleration sensor, regardless of sensor attachment site. The pattern matching method correctly distinguished ingestive chewing with an accuracy of over 0.9, regardless of the sensor site (estimated number of chews by automatic detection/true number of chews by manual detection). Placement of the sensor on the horn of the animal resulted in the most matched measurement (0.99). In contrast, the method overestimated ruminating chewing time (approximately 1.50) because noises that occurred during idling were misclassified as ruminating chewing. The provision of an additional chewing criterion for automatic detection, i.e., chewing is performed consecutively (context-based cues), improved the classification accuracy of ruminating to a range of 1.02-1.08. In conclusion, automatic detection using a pattern matching method successfully classified ingestive chewing and ruminating chewing in cattle, and a single-axis acceleration sensor was useful for acoustic monitoring at all sensor attachment sites.