In allusion to the problem of thermal fault diagnosis and location of the substation, an infrared image automatic detection method was proposed in this paper. In order to solve the target extraction problem of com- plex images, the two-dimensional Otsu threshold method was adopted to improve the region growing algorithm. On the basis of the pixel statistics that reflect the appearance characteristics of the device, the division of the structure of the target equipment was achieved by way of the rules of extreme value. According to the results of structural division and fault diagnosis criteria, the refined automatic diagnosis and positioning of the thermal faults in substation were finally realized. Experiments showed that this method can achieve thermal fault diagno- sis and location of substation well without manual intervention, which greatly improved the efficiency and accu- racy of fault diagnosis.