Road Crack Detection Using HDD LOSS and Dual Attention Module with DeepLabv3+

Crack detection and analysis on road surfaces is an essential task within the road maintenance management system. This study proposes an enhanced DeepLabv3+ algorithm that integrates Hausdorff distance Loss and Dual Attention Module (CPAM), aiming to improve the precision and accuracy of road surface crack segmentation. The road surface crack data in the dataset were trained using DeepLabv3+, DeepLabv3+ with CPAM, and DeepLabv3+ with CPAM (use HDD Loss) algorithms respectively, and the model that performed well on the test set was selected for validation on the validation set. The experimental results demonstrate that, compared to the traditional DeepLabv3+ algorithm, the enhanced DeepLabv3+ model integrating Hausdorff distance Loss and Dual Attention Module (CPAM) achieved a significant improvement in the precision of crack segmentation. The findings highlight the potential application of semantic segmentation in road crack detection and segmentation, and also provide an effective strategy for the automation of road crack detection and segmentation.