Combining automated behaviour recognition and physiological data to characterize heat tolerance and animal welfare in growing pigs

Estimating animal behaviour during heat stress (HS) is particularly insightful to monitor animal welfare but also to better understand how animals thermoregulate. The present study is a proof of concept combining computer vision to monitor animal behaviour, continuous monitoring of subcutaneous temperature and recording of ambient temperature, with the aim to study the link between behaviour and animal body temperature during HS. A total of 22 pigs were video-monitored from 8:00 to 18.00 under two contrasted conditions: HS corresponding to the tropical climate (between 20.3 °C and 27.9 °C) and Thermoneutral (TN) consisting of an indoor temperature-controlled room at 22 °C. Animal temperature (T_muscle) and ambient temperature were monitored continuously using temperature loggers. Pig postures estimated by a neural network show that animal in HS spend more time lying laterally and less time lying sternally than in TN. Moreover, in HS, the length of the lateral sequences increased with the outdoor ambient temperature. The ability of an animal to dissipate heat while lying laterally was quantified through a Heat Dissipation Coefficient (HDC), combining T_muscle and lateral lying sequence duration, and showed great individual variation. A Heat Discomfort Index (HDI) was also determined to quantify the difference in time spent lying laterally between HS and TN and could be useful as a proxy to quantify animal welfare reduction due to HS. This study demonstrates that combining image analysis to monitor animal behaviour and physiological data is an efficient tool to derive quantitative criterion to characterize animal welfare and traits related to heat tolerance.