Answers are grounded in this paper and cite their sources.
Proposed CSNet model improves detection accuracy of sealed cracks in asphalt pavements, suggesting enhanced traffic safety.
Pixel‐level detection of seal cracks on complex road surfaces is crucial for traffic safety and the structural integrity of highway bridges. This paper proposes an improved network architecture, the seal crack segmentation network (CSNet), based on HRNet‐OCR for detecting seal cracks at the pixel level on asphalt road surfaces. The CSNet is based on the Swin transformer architecture and combines a bidirectional attention mechanism with a boundary optimization module. This architecture achieves superior detection performance compared to existing methods. Key innovations include the following: (1) using a diffusion model for data augmentation to improve the learning capabilities of small samples, (2) designing a Swin transformer backbone network that better captures multiscale features by replacing the traditional HRNet, (3) proposing a dual transformer module that integrates local and global contextual information, and (4) introducing a boundary optimization module (BRM‐CRF) that improves the accuracy of edge detection. Experimental results show that the CSNet model achieves a detection accuracy of 95.75%, an F‐score of 95.67%, and an intersection‐over‐union ratio of 0.9108 on a test set of 500 images. Compared to state‐of‐the‐art semantic segmentation models, the proposed CSNet model achieves higher detection accuracy.
No takes yet. Share an insight, caveat, or question.
Hu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: