Automatic recognition method of cow ruminating behaviour based on edge computing

Timely monitoring of the ruminating behaviour of dairy cows is beneficial for obtaining relevant information on dairy cow health to predict dairy cow diseases for the first time. To date, various strategies for monitoring ruminating behaviour have been proposed, but the real-time monitoring of these strategies has not been fully realized. Based on edge computing, we proposed a real-time method to monitor the ruminating behaviour of dairy cows. In this work, a self-designed edge device was used to collect and process the three-axis acceleration signals of dairy cows in real-time, and then a rumination recognition algorithm was used to calculate the overall sliding geometric mean of the Euclidean distance between the feature sets in real-time, determine the adaptive threshold, and verify the ruminating behaviour by the sliding window. Finally, real-time recognition of the ruminating behaviour of dairy cows was completed on the edge device side, without requiring substantial calculation time and resources. The edge device uploaded cow ruminating information to the cloud in real-time every two hours, and the cloud further aggregated the ruminating information. Compared with the traditional method of uploading three-axis acceleration data, this cloud-end integrated system based on edge computing reduced the amount of uploaded data bytes by 99.9%. Our ruminating recognition has achieved the following performance values: precision (93.7%), recall (92.8%), F1-score (93.3%), specificity (97.4%) and accuracy (96.1%), indicating a good classification effect. This research provides a real-time and effective method for monitoring of cow ruminating behaviour, and the proposed system can be used in practical applications.

Automated summary

This study proposes a real-time method for monitoring dairy cow ruminating behavior based on edge computing, using a self-designed edge device for data collection and processing. By uploading data to the cloud and aggregating it, this system significantly reduces the amount of data transmitted while achieving good classification performance.