Title: Wall crack detection method based on improved YOLOv5 and U2-Net

Authors: Zujia Zheng; Kui Yang

Addresses: Office of Infrastructure, Wuhan University of Science and Technology, Hubei, Wuhan, 430081, China ' Office of Infrastructure, Wuhan University of Science and Technology, Hubei, Wuhan, 430081, China

Abstract: Wall crack detection is significant for quality inspection and project maintenance after civil construction, especially cracks in load-bearing walls and exterior walls are great safety hazards, and wall crack detection in a complex context is necessary. This paper proposes a network architecture integrating YOLOv5 and U2-Net to solve the problem of poor segmentation of crack targets in the background of large environment, optimise the problem of large computation and slow training of YOLOv5 with GhostNet network module, input the region of interest extracted by the improved YOLOv5 backbone network into U2-Net for binary classification, and fuse the segmented targets in the feature frame into the input image to improve the final classification effect and reduce the false detection rate of target detection.

Keywords: wall cracks; object detection; YOLOv5; GhostNet; U2-Net.

DOI: 10.1504/IJWMC.2023.135405

International Journal of Wireless and Mobile Computing, 2023 Vol.25 No.4, pp.362 - 367

Received: 22 Dec 2022
Accepted: 25 Feb 2023
Published online: 08 Dec 2023 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article