An image enhancement algorithm to improve road tunnel crack transfer detection

Cracks are a common disease in road transportation infrastructure, while crack detection has been a difficult task for a long time, especially for tunnels. Both training data and network models are important factors for the crack detection task using deep neural networks, but the image dataset of road tunnel surface crack is always insufficient. Because road crack images are much easier to acquire than road tunnel crack images, using road crack images for road tunnel crack detection may be a solution. To address the problem of insufficient road tunnel crack data, this paper proposed an image enhancement algorithm that is applied to road crack data. Three different object detection models are used for examining the effectiveness of crack transfer detection. Experiments in the paper show that the transfer detection model can be established between road tunnel crack data and road crack data. The detection result of YOLOv5 is the best among the three network models. The image enhancement algorithm proposed in this paper can improve the accuracy of transfer detection.