Infrared technology can effectively detect the overheating defects of power equipment. It has the characteristics of long distance, no contact, no sampling, accurate, fast, intuitive and so on. The traditional infrared artificial diagnosis of power equipment fault is time-consuming and labor-consuming, but one of the difficulties of intelligent diagnosis is whether or not to obtain the region of interest. Infrared image has the properties of concentrated intensity and low contrast. The common segmentation algorithm is used to obtain infrared image of power equipment by ROI, and the result is often over-segmentation. In order to strengthen the fault detection of the equipment, this paper is based on the fixed infrared camera installed in the electrical equipment station to monitor real-time all weather, and to film the cross-linked cable, cable head, current transformer connector, circuit breaker, dry transformer, etc. The infrared image of electrical equipment fault of 20 kinds of related equipments, such as pillar insulator, is segmented by salient detection based on robust background detection. The result shows that the image edge is connected. The background weighted contrast and the optimization effect are better than the traditional method, and the problem of image over-segmentation is avoided effectively. It is proved that the significance detection based on robust background detection is feasible for image segmentation of fault location.