To solve the problem of low efficiency and accuracy of power equipment fault diagnosis using the traditional threshold segmentation method, an intelligent algorithm, the optimized Otsu algorithm was used for threshold segmentation of infrared images for fault diagnosis. According to the shortcomings of the basic moth-flame optimization, the improved moth-flame optimization algorithm is proposed. It was applied to the infrared image segmentation. By comparing its infrared image segmentation results with those of the particle swarm optimization, biogeography-based optimization, and moth–flame optimization algorithms, it was shown that the improved algorithm is successful. A multithreshold segmentation method for infrared images through the temperature region is proposed. It can accurately determine the temperature range of each part and ensure normal operation of the equipment.