Visual detection of sand concrete cracks in railway tunnel lining mechanism based on Mask R-CNN model

The crack image of railway tunnel lining structure has complex grayscale distribution and variation characteristics, and local and global multi feature information interferes with tracking direction and boundary tracking parameters. The model’s scalability is limited, and the detection accuracy is low. Therefore, a visual detection of sand concrete cracks in railway tunnel lining mechanism based on Mask R-CNN model was proposed. Firstly, after inputting the segmented linear transformation of the sand concrete crack image, a threshold was extracted to generate a connected domain identifier. Pixel points were used as background points in the Mask R-CNN model to simultaneously detect the position of the crack area and mark pixel level edge masks, the starting point and width of the crack boundary was determined, and a cumulative visual detection method was designed. Based on the geometric characteristics and sorting results of the cracks, the crack length was calculated, and the complete crack contour was obtained. The experimental results showed that after using the method proposed in this paper, all key positions can be detected completely; After increasing the number of iterations, the detection results were less affected, indicating that its scalability had been improved and can adapt to task requirements; and therefore, it has good application value.