An Automatic Crack Detection Method for Structure Test Based on Improved Mask-RCNN

The algorithm of automatic crack identification based on computer vision has a good application prospect in aircraft structural fatigue test. However, due to the diversity of aircraft structure and the complexity of fatigue test environment, the accuracy of traditional methods for crack identification is difficult to meet the requirements. Therefore, a detection strategy based on key structure location is designed, and the model architecture and non-maximum suppression module are improved based on Mask-region convolutional neural network (Mask-RCNN), and an automatic crack identification method is proposed. This method has the characteristics of avoiding interference factors actively and having low requirements on picture quality. Meanwhile, it adopts the feature of adopting the pixel information into parameter optimization by using Mask-RCNN, which has a higher recognition accuracy rate. In component fatigue test, the identification accuracy of rivet and crack by this method is 100% and 87.5%, respectively, which has a significant advantage over the existing method. © 2021, Editorial Department of JVMD. All right reserved.