PMENet: a parallel UNet based on the fusion of multiple attention mechanisms for road crack segmentation.

The presence of road cracks significantly impacts both traffic safety and road maintenance. Therefore, accurate detection of road cracks plays a crucial role in road maintenance and management. This study focused on addressing the critical challenge of accurate road crack detection in images through the development of a novel network architecture (PMENet) based on deep learning semantic segmentation algorithms. The proposed PMENet algorithm combined parallel contextual squeeze and excitation(PCSE), multiscale feature fusion(MFF) and double residual efficient attention(DREA) modules based on the UNet structure, which improves the ability to extract global local information and crack boundary information from the model. We thoroughly evaluate the performance of the PMENet algorithm on the publicly available Crack500 dataset by the ablation experiment. We compared our method with existing approaches, demonstrating its superiority in terms of accuracy, robustness, and generalization ability. The PMENet algorithm achieves an accuracy of 96.14%, an F1-score of 84.22%, and an IoU of 62.00%. The proposed model achieves a significant improvement of 5.9% in F1-score and 5.49% in IoU compared to the UNet model. [ABSTRACT FROM AUTHOR]