The migration learning method based on inception v3 model is used to automatically detect the cracks in wind turbine blades. The full-connection layer based on BP neural network and extreme learning machine（ ELM） is designed to realize the blade state classification. The experimental results show that the convergence speed and network accuracy of the migration learning method are better than the Faster R-CNN and Yolo v3 deep learning algorithms. Moreover

for the fully connected layer with relatively simple structure

the performance similar to BP neural network can be obtained by ELM extreme learning machine（ accuracy rate 89. 3 %

recall rate91. 5 %

F1 value 0. 819）

but the training speed can be improved thousands of times. This is suitable for application scenarios with high real-time requirements.