Automatic detection algorithm for bridge crack images based on Convolutional Neural Network

In this study, a convolutional neural network (CNN)-based automatic detection algorithm for bridge crack images was proposed, aiming to improve the accuracy and efficiency of bridge structural health monitoring. Firstly, the multi-scale convolutional layer is used to extract the local features in the image to enhance the ability to identify cracks of different sizes. Then, the Residual Network (ResNet) module was introduced to solve the gradient vanishing problem in deep network training, so as to ensure that the performance of the model does not decrease but increases when the depth of the model increases. In order to further improve the detection accuracy, an attention mechanism module is designed, which can dynamically adjust the importance of the feature map of each part to make the model focus more on the crack area. In addition, the U-Net architecture is used for image segmentation to accurately locate and classify cracks while maintaining the consistency of background information. Finally, the effectiveness of the algorithm is verified by a large number of field collected image datasets, and the results show that compared with the traditional methods, the proposed algorithm has significant improvements in crack detection accuracy, recall rate and F1 score. This algorithm provides reliable technical support for bridge maintenance and helps to detect potential safety hazards in time.