Defect Detection of Fan Blades Based on Image Acquisition by Unmanned Aerial Vehicles

Traditional fan blade defect detection methods rely on manual inspection, which is inefficient and costly. Traditional detection methods use ground cameras for shooting, resulting in insufficient coverage of specific angles and areas of fan blades. By using drones, aerial photography of wind turbine blades can be achieved, which can improve the comprehensiveness of detection. This article collects a large number of fan blade images through drones, preprocesses the images, and marks them. This article improves the residual network (ResNet) model by using global maximum pooling to replace the fully connected layer and using Dropout regularization technology to reduce overfitting. The test set results showed that out of 1000 samples, 992 samples were correctly classified, and eight samples were misclassified. This article utilizes unmanned aerial vehicles (UAV) to collect images and uses an improved ResNet model to effectively improve the classification performance of fan blade defects.