In order to improve the temperature measurement accuracy of molten steel,an experimental platform based on the principle of thermal radiation temperature measurement is set up,and the infrared images of molten steel under the different temperatures are obtained.The average gray value of the image is extracted using MATLAB,and then fitting curve of the temperature and gray is obtained using the least squares method and BP neural network.Finally,the model of the infrared temperature measurement is gotten,and the molten steel temperature measurement error reaches 1%,which meets the requirement of the measurement precision.This method lays a foundation for the research on molten metal online infrared temperature measurement.