Infrared thermal imaging technology was widely used in many fields. So infrared thermal imaging technology and image processing were combined in detection of catenary thermal state. In image processing, firstly, image preprocessing was used for infrared thermal images, and then mutual information-based image template matching was used to detect and recognize the targets of infrared image. Mutual information was similarity measure criterion in image template matching, and Non-Parametric (NP) window method was to estimate the Joint Probability Distribution Function(JPDF) for underlying template. For weakness of NP window method was long computing time, Modified Non-Parametric (MNP) window method was used with a triangle distribution instead of uniform distribution in the original NP window. Experimental results show that accuracy in calculating the joint probability distribution of MNP method is higher and the image template matching time is shortened.