Detection Method of Concrete Surface Cracks in Bridge Structures Using DeepLabV3+ Enhanced with Mona

Bridge crack detection technology is gradually transitioning from manual surveying to automated detection using mobile platforms such as drones and unmanned boats. However, this new detection mode imposes high demands on algorithm real-time performance and faces the challenge of class imbalance in images. To address these issues, this paper designs a multi-strategy data preprocessing process and proposes the MDLab lightweight semantic segmentation model, which incorporates a Mona-Integrated Multi-scale Feature Module and a Multichannel Feature Fusion Module. With only 23.9M parameters, the model achieves a mean Intersection over Union (mIoU) of 88.27%, which is 1.58% higher than DeepLabV3+, and processes a single image in 22ms, a 45ms reduction compared to U-Net. The results demonstrate that the proposed model achieves high segmentation accuracy while maintaining a lightweight architecture compared to mainstream models.