Methodology for data processing and analysis techniques of infrared video thermography used to measure cattle temperature in real time

Abstract

Core body temperature is a measurement used to evaluate and diagnose the health status of cattle, with rectal temperature measured using an indwelling probe being the preferred method. This procedure can be stressful to animals, potentially generating inaccurate results and is impractical. Infrared thermography (IRT) of the ocular region of the eye in cattle may be a less stressful and safer alternative. The objective of the present study was to develop a method to measure the temperature of cattle using IRT under commercial settings. One-hundred and twenty cattle, of mixed breeds were used in Experiment 1 and were recorded by 3 different IRT cameras placed on the race in a commercial abattoir. Experiment 2 used 52 cattle of mixed breeds that were recorded by 4 different IRT cameras at the same sample place and distance from the cattle crush on a farm. Animals in Experiment 2, were caught in the crush where IRT data and rectal temperature was recorded. Under commercial abattoir conditions, IRT data shows large variability ranging from 23.1° to 47.5 °C, calling for extensive data processing. When the eye was within the region of interest (ROI) under commercial conditions it reflected the highest temperature in comparison to the skin or facilities. The maximum temperature within the ROI was extracted from the frames containing the ocular region of the eye (raw data) and then these data processed, calculating the rolling median and quantiles. The correlation between eye IRT and rectal temperature is lower when using unprocessed raw data (r = 0.19; P > 0.05) compared to using quantiles or the rolling median (r = 0.43; P < 0.05) due to the presence of outliers. The frame rate, image resolution and accuracy of the cameras also affected the correlation between eye IRT and rectal temperature. This resulted in some data processing methods that worked well for some cameras but not for others. Reducing the length of time animals were within the ROI produced a reduction in the strength of the correlation between IRT eye and rectal temperature. However, some cameras such as the FLIR A310 were less affected by the length of time. We demonstrate that a 1-second rolling median of the maximum and the 99th quantile is an appropriate methodology to analyse IRT data collected at a high frequency using video cameras. Data from video IRT cameras require extensive processing before it can be used to measure body temperature in cattle under commercial conditions.