The invention discloses a method for diagnosing defects of an insulator image based on the insulator image, and the method comprises the following steps of carrying out the Gaussian low-pass filteringand denoising preprocessing of the insulator image, and removing noise pixels in the image; carrying out color channel weighted superpixel clustering on the denoised image, and segmenting the image into a certain superpixel; and dividing the processed images into a training set and a test set, adding labels, and selecting a support vector machine algorithm to classify the images to obtain a prediction result. By utilizing the method, the recognition speed of the insulator defects can be fully improved on the premise that the recognition rate is not reduced, and the method has a certain valuefor improving the efficiency of electric power inspection and reducing the cost and also has a certain significance for guaranteeing the safe operation of a power grid.