Electric power equipment temperature is closely related with its working condition. The current mainstream infrared thermal imagers are expensive, complicated to operate and weak in spatial positioning. Infrared and visible image fusion has a promising application prospect in power system fault location, anomaly monitoring and so on. Visible light radiation temperature measurement has many application cases in the high temperature field. This paper puts forward a method based on image processing and machine learning, and successfully applies visible light temperature measurement technology to the low temperature field such as the temperature detection of power equipment. We used copper plates as the research object, and established a library of visible images under different temperature and light conditions. Four kinds of machine learning algorithms were used to build the temperature prediction model by extracting the gray distribution features from images. We select two algorithms which have good performance in time complexity and prediction accuracy. The average absolute error of predicting temperature is only about 1.5℃. We also have performed Retinex processing on all images to eliminate the interference of different lighting intensity on the grayscale features. After training and calculation, it was found that the average absolute error is reduced to 1.309℃ with the same algorithms, which has a better prediction accuracy.