Tunnel Crack Segmentation Algorithm Based on Feature Enhancement

Tunnel cracks, as one of the early indicators of tunnel damage, have a significant impact on the safe operation of tunnels. However, due to the complex lighting conditions, background noise, and diverse nature of cracks inside tunnels, traditional image segmentation algorithms often struggle to achi...

Full description

Saved in:
Bibliographic Details
Main Authors: Lihua Feng, Aijun Yao, An Huang
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/10971940/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850170878680104960
author Lihua Feng
Aijun Yao
An Huang
author_facet Lihua Feng
Aijun Yao
An Huang
author_sort Lihua Feng
collection DOAJ
description Tunnel cracks, as one of the early indicators of tunnel damage, have a significant impact on the safe operation of tunnels. However, due to the complex lighting conditions, background noise, and diverse nature of cracks inside tunnels, traditional image segmentation algorithms often struggle to achieve sufficient accuracy in crack segmentation tasks. To address these challenges, this paper proposes a tunnel crack detection method. Firstly, a novel Dynamic Feature Enhancement Network (DFEN) module is designed to extract preliminary features using a feature extractor and selectively enhance feature representations through a gating mechanism. Additionally, a new Multi-Head Interaction Learning (MHIL) is introduced, facilitating feature sharing by incorporating information exchange across different attention heads, thereby improving feature representation capabilities. Lastly, the Adaptive Switchable Atrous Convolution (ASAC) module is introduced, combining the advantages of adaptive convolution and deformable convolution while incorporating Switchable Atrous Convolution (SAC) to enhance multi-scale feature capturing capabilities. Ablation experiments, through both qualitative and quantitative evaluations, demonstrate the effectiveness of the proposed method. Comparative analysis with existing semantic segmentation methods further confirms the superiority of the proposed method in tunnel crack detection.
format Article
id doaj-art-495fa2a9370f4bbf9b0a7dda496da6de
institution OA Journals
issn 2169-3536
language English
publishDate 2025-01-01
publisher IEEE
record_format Article
series IEEE Access
spelling doaj-art-495fa2a9370f4bbf9b0a7dda496da6de2025-08-20T02:20:23ZengIEEEIEEE Access2169-35362025-01-0113706507066210.1109/ACCESS.2025.356292610971940Tunnel Crack Segmentation Algorithm Based on Feature EnhancementLihua Feng0https://orcid.org/0009-0008-4890-2771Aijun Yao1An Huang2College of Architecture and Civil Engineering, Beijing University of Technology, Beijing, ChinaCollege of Architecture and Civil Engineering, Beijing University of Technology, Beijing, ChinaCollege of Architecture and Civil Engineering, Beijing University of Technology, Beijing, ChinaTunnel cracks, as one of the early indicators of tunnel damage, have a significant impact on the safe operation of tunnels. However, due to the complex lighting conditions, background noise, and diverse nature of cracks inside tunnels, traditional image segmentation algorithms often struggle to achieve sufficient accuracy in crack segmentation tasks. To address these challenges, this paper proposes a tunnel crack detection method. Firstly, a novel Dynamic Feature Enhancement Network (DFEN) module is designed to extract preliminary features using a feature extractor and selectively enhance feature representations through a gating mechanism. Additionally, a new Multi-Head Interaction Learning (MHIL) is introduced, facilitating feature sharing by incorporating information exchange across different attention heads, thereby improving feature representation capabilities. Lastly, the Adaptive Switchable Atrous Convolution (ASAC) module is introduced, combining the advantages of adaptive convolution and deformable convolution while incorporating Switchable Atrous Convolution (SAC) to enhance multi-scale feature capturing capabilities. Ablation experiments, through both qualitative and quantitative evaluations, demonstrate the effectiveness of the proposed method. Comparative analysis with existing semantic segmentation methods further confirms the superiority of the proposed method in tunnel crack detection.https://ieeexplore.ieee.org/document/10971940/Crack segmentationdynamic feature enhancement networkmulti-head interaction learningadaptive switchable atrous convolution
spellingShingle Lihua Feng
Aijun Yao
An Huang
Tunnel Crack Segmentation Algorithm Based on Feature Enhancement
IEEE Access
Crack segmentation
dynamic feature enhancement network
multi-head interaction learning
adaptive switchable atrous convolution
title Tunnel Crack Segmentation Algorithm Based on Feature Enhancement
title_full Tunnel Crack Segmentation Algorithm Based on Feature Enhancement
title_fullStr Tunnel Crack Segmentation Algorithm Based on Feature Enhancement
title_full_unstemmed Tunnel Crack Segmentation Algorithm Based on Feature Enhancement
title_short Tunnel Crack Segmentation Algorithm Based on Feature Enhancement
title_sort tunnel crack segmentation algorithm based on feature enhancement
topic Crack segmentation
dynamic feature enhancement network
multi-head interaction learning
adaptive switchable atrous convolution
url https://ieeexplore.ieee.org/document/10971940/
work_keys_str_mv AT lihuafeng tunnelcracksegmentationalgorithmbasedonfeatureenhancement
AT aijunyao tunnelcracksegmentationalgorithmbasedonfeatureenhancement
AT anhuang tunnelcracksegmentationalgorithmbasedonfeatureenhancement