Pixel level algorithm and application of bridge surface cracks based on BSCS-Net

To solve the problems such as low recognition accuracy and poor generalization in the current intelligent detection of bridge surface cracks, this paper proposes a pixel-level intelligent detection algorithm for bridge surface cracks (Bridge Surface Crack Segmentation Network, BSCS-Net). Firstly, based on the images of the surface cracks of the bridge collected by the truss-type bridge inspection vehicle, a sample data set is manually constructed; Then, through a series of ablation experiments, it was proved that integrating the DSM and CA modules could effectively improve the algorithm's ability to understand the global background and crack features of the crack images; Finally, the crack segmentation effect is compared with the existing semantic segmentation algorithms using the sample data set. The results show that: On the sample dataset constructed in this paper, the accuracy rate, recall rate, F-measure and intersection-union ratio of the proposed algorithm are 92.36%, 78.01%, 84.58% and 73.28% respectively, and the processing speed is 22.19 frames per second. All are superior to U-Net, DeepLabv3+, SegFormer and DC-Unet, proving that the proposed algorithm has the optimal detection accuracy and speed.