Wind turbines capture the kinetic energy produced by the wind, with the blades being the component most susceptible to damage. Unplanned stops result in significant losses, which highlights the need to detect these failures early. As a step in the preventative maintenance procedure, hundreds of color photographs of the blades are taken for subsequent analysis by an expert. In this work, we present a method to highlight superficial damages in wind turbine blades using a classification and localization process. A new dataset was created using images of wind blades, each divided into uniform slices and labeled by an expert according to the type of fault identified. Then, we apply class balancing and data augmentation methods prior to fine-tuning a general-propose pre-trained deep convolutional neural network model. The best model was used to classify and locate damages in the wind blades images. As a result, we obtained an overall precision of 96.1%, accuracy of 97% and recall of 94.5% when classifying the presence of damages. We show that our method can be integrated into the monitoring of wind blade tasks, helping the specialist to highlight and identify images containing damage.