· For the first time, the deep learning algorithm and photometric stereo algorithm 3D reconstruction were applied to the detection of hydrogen fuel bipolar plates, filling the international and domestic gaps;

· Use multi-angle light source imaging data, use photometric stereo algorithm for 3D reconstruction, and combine 2D and 3D detection methods to improve the recognition of defects;

· The deep learning algorithm adopted solves the problems of positioning, detection and classification of bipolar plate defects;

· The detection system has a powerful self-learning function. With the continuous operation of the software, the defect detection rate will continue to increase;