ABSTRACT The objective of the present study was to validate the accuracy of algorithms, implemented in the currently available RumiWatch Converter (RWC) version V0.7.4.5 of the RumiWatch System (RWS), for the classification of behavioural characteristics from jaw and head movements which are monitored by a noseband halter comprising a pressure sensor and a triaxial accelerometer. The accurate classification of behavioural characteristics in different time resolutions is critical for the usage of the RWS for scientific and practical purposes as chewing behaviour provides essential indicators for the assessment of diet adequacy in dairy cows. To validate the RWC V0.7.4.5 classification accuracy for behavioural characteristics of rumination, eating, drinking, other activity and ruminating chews per bolus by direct observation as reference method, 14 dairy cows participated in the trial. Concordance between the consolidated 1-min and 1-h classification results was assessed. The RWC V0.7.4.5 classified only rumination and ruminating chews per bolus precisely, whereas an algorithm optimisation for the classification of eating, drinking and other activity is required. Additionally, classification results from the 1-min and 1-h time summaries were not in agreement with each other except for rumination.

中文翻译：

---

验证 RumiWatch Converter V0.7.4.5 分类准确性，用于自动监测奶牛的行为特征

摘要 本研究的目的是验证算法的准确性，该算法在 RumiWatch 系统（RWS）的当前可用的 RumiWatch 转换器（RWC）版本 V0.7.4.5 中实现，用于对来自下巴和头部运动的行为特征进行分类由一个由压力传感器和三轴加速度计组成的鼻带挂脖监控。不同时间分辨率下行为特征的准确分类对于将 RWS 用于科学和实用目的至关重要，因为咀嚼行为为评估奶牛的饮食充足性提供了重要指标。为了通过直接观察作为参考方法，验证 RWC V0.7.4.5 对反刍、进食、饮水、其他活动和每团反刍咀嚼行为特征的分类准确性，14头奶牛参加了试验。评估了综合 1 分钟和 1 小时分类结果之间的一致性。RWC V0.7.4.5 仅对每次推注的反刍和反刍咀嚼进行精确分类，而需要对饮食和其他活动进行分类的算法优化。此外，除了反刍之外，1 分钟和 1 小时时间总结的分类结果彼此不一致。