In order to quickly find the insulator defects in the image and reduce the probability of power accidents, an insulator image segmentation algorithm is designed based on the lightweight Fast-Unet network. The method comprises the following steps of preprocessing an aerial image of an insulator to realize reconstruction and normalization; calculating the element symbiosis probability by converting image elements so as to obtain the color, texture and shape characteristics of the image; refining the channel space information characteristic value to form an initial network structure and generate a segmentation result dendrogram. By calculating the modularity and relative modularity of the image, a lightweight Fast-Unet network segmentation model is established. The experimental results show that the average comprehensive quality of the proposed algorithm is 0.72 in noise-free images, and 0.57 and 0.46 in simple and complex background images, respectively. Therefore, denoising has a positive effect on the quality of image segmentation.