Breathing Pattern Analysis in Cattle Using Infrared Thermography and Computer Vision

Abstract

Breathing patterns can be considered a vital sign providing health information. Infrared thermography is used to evaluate breathing patterns because it is non-invasive. Our study used not only sequence temperature data but also RGB images to gain breathing patterns in cattle. Mask R-CNN was used to detect the ROI (region of interest, nose) in the cattle RGB images. Mask segmentation from the ROI detection was applied to the corresponding temperature data. Finally, to visualize the breathing pattern, we calculated the temperature values in the ROI by averaging all temperature values in the ROI. The results in this study show 76% accuracy with Mask R-CNN in detecting cattle noses. With respect to the temperature calculation methods, the averaging method showed the most appropriate breathing pattern compared to other methods (maximum temperature in the ROI and integrating all temperature values in the ROI). Finally, we compared the breathing pattern from the averaging method and that from the thermal image observation and found them to be highly correlated (R2 = 0.91). This method is not labor-intensive, can handle big data, and is accurate. In addition, we expect that the characteristics of the method might enable the analysis of temperature data from various angles.

Keywords

• breathing pattern
• infrared thermography
• cattle health and welfare
• computer vision
• machine learning