Respiratory rate detection of dairy cows based on infrared thermography in head movement scenarios

The respiratory rate (RR) of an individual cow is commonly used as a physiological indicator to monitor its heat stress and health conditions. The monitoring of temperature changes in dairy cow nostrils using infrared thermography (IRT) is an effective technique for monitoring the RRs of cows. However, the high-resolution requirements for thermal images and the effect of abnormal respiratory curve fluctuations due to cow head movements on RR detection limit the applicability of this technique. This study aimed to develop an RR detection method that is based on low-resolution thermal images and incorporated a curve fusion strategy to mitigate the effects of head movements. In this method, the YOLOv8n-Pose network is first utilized to identify nostrils as key points, which addressed the challenge of nostril region localization at low resolution. Second, the random forest algorithm is employed to establish a mapping relationship between the temperature of each pixel and the corresponding RGB color space values. Finally, a fusion strategy for the temperature curves of the two nostrils is implemented to address the issue of abnormal temperature curve fluctuations that are caused by head movements. To evaluate the proposed method, a dataset of 246 thermal videos of cow faces was constructed. The results showed that the correlation analysis between the predicted RR and manually observed results yielded an R² of 0.92 and a root mean squared error (RMSE) of 3.53 breaths per minute (bpm), and the average accuracy of RR detection reached 96.3 %. This study demonstrated that the integration of IRT and deep learning provides a foundation for the deployment of automatic RR detection in dairy farming environments.