Abstract: Body temperature is one of the most important indicators of disease diagnosis in pigs. In order to reduce the manpower and material resources used in the measurement of traditional pigs' temperature methods and decrease the risk of pigs' stress and cross-infection between humans and pigs, the infrared thermal imager (Fluke Ti27) was used to acquire images of sows infrared heat radiation. The deep learning target detection network YOLOv3 was used to train and predict the dataset to accurately identify and locate the ear root of sow. The ear root part of the sow was selected as the best measurement part and with the temperature information of the ear root part in the thermal radiation picture obtained by the Fluke software (Fluke Connect SmartView), the relationships between the sow body temperature and the ambient temperature, the ambient humidity, the light intensity, the infrared temperature of the ear root were analyzed so that the multiple linear regression model with sow body temperature as the dependent variable and other variables as independent variables was established and the multiple linear regression function was used to optimally fit the sow body temperature. Using this model to estimate the data of the test set, the results showed that: under different environmental conditions, the maximum error between the fitted pig body temperature and the actual pig body temperature was 3.06%, and the average absolute error was 1.41%. The body temperature fitting is accurate, and the fitting error basically meets the pig breeding industry requirements. This method can be used as a non-contact measurement of pig body temperature in pig production, which improves the accuracy and efficiency of temperature measurement and has a good prospect.

Key words: sow, infrared thermography, body temperature measurement, multiple linear regression