Abstract: In order to explore the relationship between the infrared temperature and the true temperature of the laying hens, the explanatory variables were added to improve the degree of interpretation of the relationship between the two by the prediction model, and the collinearity between the explanatory variables was eliminated to improve the accuracy of the prediction model. In the experiment, the infrared image of the laying hen taken by the infrared thermal imager was divided into several regions of interest; the infrared temperature and ambient temperature of the region of interest were taken as the explanatory variables of the inversion model, and the temperature under the wings of the laying hens was used as the dependent variable to establish a multiple linear regression prediction model of laying hens body temperature. At the same time, ridge regression was used to optimize the prediction model, and on the same test set, it was compared with the prediction model under the BP neural network architecture to analyze the average relative error of different models. The results showed that the average relative error of multiple linear regression model was 1.06%; the average relative error of the ridge regression model was 0.71%, and the average relative error of the BP neural network model was 0.28%.When the body temperature of the laying hen was between 41 ℃ and 42 ℃, the three models could accurately predict the real body temperature of the laying hens within the allowable error range. When the body temperature of the laying hens exceeded 42 ℃, the error between the ridge regression model and the BP neural network model were mostly distributed between 0-1%, which met the accuracy requirements. The prediction accuracy of BP neural network model was the best, while the multiple linear regression model differed greatly from the real temperature, so it could not accurately predict the real temperature of laying hens. Compared with the multiple linear regression model, the ridge regression model and the BP neural network model could eliminate the collinearity between explanatory variables, improve the stability of the model, and make the prediction results more accurate.