Bridge crack detection algorithm designed based on YOLOv8

Abstract

Due to the complexity of the operating environment and the influence of natural factors, bridges are prone to various forms of damage, including cracks. However, traditional bridge detection methods often encounter challenges in terms of low detection accuracy and high computational resource consumption. The practical significance of accurate bridge crack inspection to society can be summarized as follows: ensuring bridge safety, preventing major accidents, maintaining bridge structural health, optimizing bridge management decisions, promoting scientific and technological progress, and enhancing public trust. The practical significance of accurate bridge crack inspection extends beyond the safety and stability of the bridge itself. It also relates to the safety of people's lives and property, as well as the harmony and stability of society. This study presents a bridge crack detection algorithm tailored around the YOLOv8 framework. Initially, the SPPF_UniRepLk module is incorporated into the algorithm's backbone network, aiming to bolster its capacity to capture and extract pertinent image features. Additionally, to further grasp the global dependencies between feature maps, a Global Channel Spatial Attention (GCSA) mechanism is introduced, which enhances the algorithm's sensitivity to global contextual information. Finally, in the neck network component of the algorithm, the Coordattention-Concat module is utilized to achieve the integration and refinement of multi-source features through nonlinear transformations and feature reweighting techniques, thereby significantly elevating the overall performance of the algorithm. The experimental outcomes demonstrate that the proposed bridge crack detection algorithm designed based on YOLOv8 achieves a mAP50–95 of 72.1 %, which is capable of accurately detecting cracks.