Abstract
With the gradual increase of highway coverage, the frequency of road cracks also increases, which brings a series of security risks. It is necessary to detect road cracks, but the traditional detection method is inefficient and unsafe. In this paper, deep learning is used to detect road cracks, and an improved model BiTrans-YOLOv5 is proposed. We add Swin Transformer to YOLOv5s to replace the original C3 module, and explore the performance of Transformer in the field of road crack detection. We also change the original PANet of YOLOv5s into a bidirectional feature pyramid network (BIFPN), which can detect small targets more accurately. Experiments on the data set Road Damage show that BiTrans-YOLOv5 has improved in Precision, Recall, F1 score and mAP@0.5 compared with YOLOv5s, among which mAP@0.5 has improved by 5.4%. It is proved that BiTrans-YOLOv5 has better performance in road detection projects.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have