Automatic defect detection and segmentation of tunnel surface using modified Mask R-CNN

The detection of tunnel surface defects is the very important part to ensure tunnel safety. Traditional tunnel detection mainly relies on naked-eye inspection, which is time-consuming and error-prone. In the past few years, many defect detection methods based on computer vision have been introduced. However, these methods with manual feature extraction do not perform well in detecting tunnel defects due to the complicated background of tunnel surfaces. To address these problems, this paper proposes a novel tunnel defect inspection method based on the Mask R-CNN. To improve the accuracy of the network, we endow it with a path augmentation feature pyramid network (PAFPN) and an edge detection branch. These improvements are easy to implement, with subtle extra memory and computational overhead. In this paper, we perform a detailed study of the PAFPN and the edge detection branch, and the experiment results show their robustness and accuracy in tunnel defect detection and segmentation.