Abstract: Based on the principle of radiation temperature measurement, we introduced the theory of infrared temperature measurement and sparse decomposition denoising principle in order to improve the molten steel temperature measuring accuracy and remove the noise of infrared image of molten steel. Having set up the experimental platform, we obtained different steel infrared images at different temperatures and experimental conditions, and used the sparse decomposition denoising for molten steel infrared image denoising processing.Compared with the Wiener filter denoising, the noise in the molten steel infrared image can be better removed by sparse decomposition and the peak signal to noise ratio increases by 10 dB.