The Surface Damage Identifications of Wind Turbine Blades Based on ResNet50 Algorithm

Abstract

Deep learning was widely used in the field of image recognitions to solve a variety of engineering problems. Aiming at the problems of high cost and low efficiency caused by traditional detection of wind turbine blades, this paper applied a deep learning classification method, ResNet50 algorithm, to identifying the surface damage of wind turbine blades based on UAV machine vision. In the study, DJI UAV were first used to collect image data, and the training set and test set were constructed after segmentations. The performance of both ResNet50 classifier and AlexNet classifier are compared to show the advantages of the ResNet50 classifier. With the same sample data and training parameters, the ResNet50 classifier achieves an average classification accuracy rate of 95.58%, which is higher than the average classification accuracy rate of the AlexNet classifier of 94.19%, which verifies the effectiveness of this method.