To enhance the visual effect of the infrared thermal image of electrical equipment and accurately detect its operating state, an image segmentation method based on a new threshold algorithm is proposed. First, the method performs Fourier filtering on the original image to form an automatic gradient graph. Then, for each specific type of fault distribution, the line model with N adjacent points is fitted to calculate the slope difference to find the optimal threshold for different types of fault regions. Finally, by morphological iterative etching, the target area is separated from the noise spots to obtain a clear segmentation image. This method is suitable for various fault types, which only requires calibration of N and determination of different segmentation cases, with others being processed automatically. The results show that the segmentation accuracy of this method is 85%, and the error rate is 0.0182%. The effectiveness and versatility of the proposed method are verified by using different types of infrared thermal fault images.