Detail-aware segmentation of pavement cracks through multimodal and multiscale feature extraction

Objectives Road surface cracks are a form of pavement deterioration. Pixel-level segmentation distinguishes crack pixels from non-crack pixels, supporting maintenance and traffic safety. However, single-modality methods cannot fully exploit complementary imaging information, resulting in missed detections, fragmented segmentation, and inaccurate boundaries under complex backgrounds, indistinct edges, and low contrast.Methods This paper proposes DAMP-UNet, a pavement crack segmentation network based on multimodal fusion and detail enhancement. Images from different modalities are fused along the channel dimension. Lightweight Token-MLP (Tok-MLP) modules in the encoder and decoder enhance multi-level spatial feature interactions. A Multi-scale Contextual Feature Extraction (MSCFE) module at the encoder end aggregates crack-related context across receptive fields. A Detail-aware Multi-scale Refinement Module (DMRM) in the decoder performs local structural enhancement, channel recalibration, and residual refinement on upsampled features before skip-connection fusion.Findings On a publicly available pavement crack dataset, DAMP-UNet improves Dice and IoU by 2.6 and 4.1 percentage points, respectively, over standard U-Net, and by 3.1 and 5.2 percentage points, respectively, over DeepCrack.Novelty The design combines multimodal fusion, Tok-MLP-based spatial feature interaction, multi-scale context extraction, and detail-aware refinement to better reconstruct fine cracks, indistinct boundaries, and discontinuous structures in complex pavement scenes.

Authors

Institutions

Publication Details

Journal
International Journal of Pavement Engineering
Published
2026-11-18
DOI
https://doi.org/10.1080/10298436.2026.2763737
Primary Topic
Infrastructure Maintenance and Monitoring
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Detail-aware segmentation of pavement cracks through multimodal and multiscale feature extraction

Kaige Xu, Yuemin Zhu, Xiaoming Sun, Yongji Chen et al.
5 links
International Journal of Pavement Engineering
Infrastructure Maintenance and Monitoring
article

Detail-aware segmentation of pavement cracks through multimodal and multiscale feature extraction

Kaige Xu, Yuemin Zhu, Xiaoming Sun, Yongji Chen, Bochao Su, Kun Sun, Shilin Li, Yan Chen
article en

Abstract

Objectives Road surface cracks are a form of pavement deterioration. Pixel-level segmentation distinguishes crack pixels from non-crack pixels, supporting maintenance and traffic safety. However, single-modality methods cannot fully exploit complementary imaging information, resulting in missed detections, fragmented segmentation, and inaccurate boundaries under complex backgrounds, indistinct edges, and low contrast.Methods This paper proposes DAMP-UNet, a pavement crack segmentation network based on multimodal fusion and detail enhancement. Images from different modalities are fused along the channel dimension. Lightweight Token-MLP (Tok-MLP) modules in the encoder and decoder enhance multi-level spatial feature interactions. A Multi-scale Contextual Feature Extraction (MSCFE) module at the encoder end aggregates crack-related context across receptive fields. A Detail-aware Multi-scale Refinement Module (DMRM) in the decoder performs local structural enhancement, channel recalibration, and residual refinement on upsampled features before skip-connection fusion.Findings On a publicly available pavement crack dataset, DAMP-UNet improves Dice and IoU by 2.6 and 4.1 percentage points, respectively, over standard U-Net, and by 3.1 and 5.2 percentage points, respectively, over DeepCrack.Novelty The design combines multimodal fusion, Tok-MLP-based spatial feature interaction, multi-scale context extraction, and detail-aware refinement to better reconstruct fine cracks, indistinct boundaries, and discontinuous structures in complex pavement scenes.

International Journal of Pavement EngineeringVol. 27(1)
Harbin University of Science and Technology (CN), Université Claude Bernard Lyon 1 (FR), Centre National de la Recherche Scientifique (FR), Inserm (FR), Shenzhen Polytechnic University (CN), Centre de Recherche en Acquisition et Traitement de l'Image pour la Santé (FR), Institut National des Sciences Appliquées de Lyon (FR)
Shenzhen Polytechnic, Natural Science Foundation of Heilongjiang Province
Industry, innovation and infrastructure
Infrastructure Maintenance and Monitoring
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon