Infrared image recognition of power equipment is a necessary prerequisite and a key step to using the infrared image for equipment defect detection, while one of the difficulties is to solve the problem of low accuracy of equipment recognition, which is caused by the difficulty of extracting equipment feature quantity caused by different rotation Angle and scale of equipment image. In this paper, 12 kinds of common power equipment, such as lightning arresters, current transformers, voltage transformers, circuit breakers, and insulators, are taken as the research objects. An improved HOG-SVM algorithm for infrared image recognition of power equipment with rotation invariance is proposed. Experimental results show that the proposed method can effectively compensate for the shortcoming of the HOG algorithm without rotation invariance, and the average recognition accuracy of 12 types of equipment reaches 92.14%, which significantly improves the recognition accuracy of infrared images of power equipment, and verifies the effectiveness and practicability of the proposed algorithm.