In order to quickly analyze and identify the massive images collected by the intelligent inspection system in the substation, an image classification model combining deep learning and support vector machine(SVM) is proposed. First, the original data collected was expanded by means of rotation, folding, etc. The expanded images were then randomly divided into training sets and test sets according to the types, and the parameters in the convolutional neural network (CNN) were improved by actual data, and the training set image features were extracted on the trained CNN models. Finally, the SVM classifier was trained by using the depth features of the training set image, and the test set image was classified and verified. The experiment uses 8000 pictures collected by the inspection robot in a substation to verify the accuracy of the model. The results show that the model has strong classification performance.