This work proposes an automatic method for the classification of partial discharge signals in insulation systems of high voltage electric generators, using a computer vision technique based on the Bag of Features algorithm in a database composed of images of Phase-Resolved Partial Discharge of corona, internal, and surface discharges in the electrical insulation of the stator bars of high voltage generators. The algorithm was implemented in a computational environment, combining the extraction of characteristics from images through the Bag of Features technique with the supervised learning process, Support Vector Machine, in the training actions and classification of signals of the automated routine. The proposed method achieved a very high classification accuracy for the three types of partial discharge signals. The results demonstrate the feasibility of integrating this approach into predictive maintenance systems for high-voltage electrical equipment, in this case, high-voltage electric machines, enhancing operational reliability and reducing costs in maintenance processes.