Aiming at the problem of small image area occupied by insulators and difficulty in segmentation and positioning, a simple non-iterative clustering (SNIC) combining edge information and convolutional neural network insulator image segmentation method is proposed. On one hand, the edge information extracted by the anisotropic diffusion filter is introduced into the SNIC algorithm to avoid the appearance of small and semantically difficult superpixels; On the other hand, based on the multi-receptive field convolution kernel, a convolutional neural network model is created to extract richer superpixel feature information to achieve accurate classification of superpixels. The experimental results based on the CPLID dataset show that the SNIC algorithm with fusion edge information improves the image edge recall rate by 6% compared with the traditional SNIC algorithm, and the classification accuracy of the multi-receptive field convolutional neural network is improved by 3.8% compared with AlexNet. Compared with other image segmentation methods, the proposed algorithm has better segmentation and positioning effect on insulators.