Different Scale Crack Contour Extraction Techniques for Bridges Incorporating Improved YOLOv5 and GCNet

Bridge cracks, as the main form of damage to bridge structures, have a significant impact on the safety and service life of bridges. Aiming at the lack of accuracy and efficiency of the traditional detection methods and the existing deep learning-based detection methods, this study proposes a contour extraction model for bridge cracks at different scales based on the deep learning algorithm. This can improve the accuracy of crack detection and the precision of contour extraction. It can also provide technical support for bridge health monitoring. In this study, the attention mechanism and double-cone feature fusion are adopted to improve the target detection algorithm, enhancing the algorithm’s ability to extract crack features. Through multi-stage processing, crack target detection is carried out first. Then, the crack contour is extracted and a smart bridge crack contour extraction model is established. The results showed that the segmentation efficiency of the research model for crack images of different sizes was 42 frames per second. The intersection-union ratio coefficient tested on the dataset was 0.93, and the pixel accuracy value was 0.99. However, in real complex environments, the accuracy, recall, and average precision of the model were tested with average values of 50-90, which were 90%, 0.89, and 0.83, respectively. The model can accurately detect and extract crack profiles at different scales. It provides efficient and accurate technical support for bridge health monitoring. This helps to extend the life of bridges and ensure traffic safety.