Pavement crack detection based on deep learning

Abstract

Effective and timely crack identification is essential to repair and limit road aging. So far, most crack detection follows manual testing rather than automatic detection based on image, which make the whole process is expensive and time-consuming. In this study, we proposed a deep learning network, that use YOLO v3 and adaptive spatial feature fusion (ASFF) strategies to enhance, label, and learn crack images. Realizing the precise classification and identification of pavement cracks. At the same time, the optimization method for identifying cracks is proposed. We chose 2000 images used in the training, which obtained from the data. 500 road images were used in the testing. The feasibility of the proposed detector is measured by precision and speed. The successful application of this study will help to identify abnormal roads which require emergency repairs, thereby improving the performance of monitoring systems for civil infrastructure.