Intelligent recognition of cold-region bridge defects based on MPRNet and improved YOLOv11
Year: 2026
Type: article
Abstract: The phenomenon of uneven ice and snow coverage and lighting under complex meteorological conditions in cold regions seriously deteriorates the quality of bridge surface images, resulting in the degradation of defect features during the acquisition process, posing a significant challenge to the recognition accuracy of existing intelligent detection models. Therefore, this study developed an intelligent recognition method for bridge defects in cold regions that integrates image restoration and object detection. This method uses a multi-stage progressive image restoration network combined with a contrast limited adaptive histogram equalization algorithm to form the preprocessing front-end, which implements gradual denoising and edge enhancement on low-quality images. On this basis, an improved YOLOv11 detector integrating adaptive downsampling and multi head self attention mechanism was designed for precise localization and classification of multiple types of defects in restored images. The results of the experiments show that the proposed algorithm achieves an average precision of 0.905 in detecting bridge defects. Regarding the four typical defects of cracks, concrete spalling, steel corrosion, and damage to the protective layer, the accuracy of crack identification and the recall rate of protective layer damage reached 79.81 and 73.25%, respectively. Moreover, the success rate of system startup and single image detection accuracy remained above 90% under low temperature conditions of − 20 ℃. The results demonstrate that this method can provide reliable intelligent identification of bridge defects in cold regions and can be used as an effective technical solution for the safety inspection and intelligent operation.
Source: Discover Applied Sciences
Author: Meng Sun
Language: English
Subfield: Civil and Structural Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7235017884
