Vision-Based Defect Detection for Maintenance Inspection of Slope Concrete Support Structures

Abstract Slope concrete support structures are continuously subjected to geological actions, weathering and erosion, and service loads, making their surfaces susceptible to defects such as cracks and spalling. Traditional manual inspection is inefficient, costly, and highly subjective, making it difficult to meet the demands for rapid, large-scale, and standardized inspection of high and steep slopes. To address feature loss and insufficient detection accuracy caused by the complex morphology of cracks and the irregular boundaries of spalling, this study proposes SDIN-YOLO (Space-to-depth convolution, DySample, Inner Complete Intersection over Union, and Normalized Wasserstein Distance-YOLO), a YOLO26-based detection model integrating SPDConv (S), DySample (D), Inner-CIoU (I), and NWD (N) through three optimizations: (1) a SPDConv module is introduced into the backbone network to reduce feature loss during downsampling; (2) the DySample upsampler, which is based on dynamic point sampling, is introduced into the neck network to improve multiscale feature reconstruction; and (3) the Inner-CIoU loss, which combines an auxiliary bounding-box mechanism with CIoU, is adopted and further integrated with the NWD to form a joint loss function, thereby improving bounding-box localization accuracy. The model is trained and evaluated on images annotated with crack and spalling defects, and its detection performance and the effectiveness of each proposed improvement are validated through comparative and ablation experiments. Experimental results showed that SDIN-YOLO achieves a precision of 89.7%, an F1-score of 74.5%, an mean average precision ( mAP 50 ) of 73.5%, and an mAP 50 – 95 of 52.7%, representing improvements of 9.4, 2.0, 1.6, and 5.5 percentage points, respectively, over the baseline YOLO26n model. Experiments under simulated foggy conditions further demonstrated that the model exhibits a certain degree of adaptability to foggy environments. The proposed method can provide defect category and location information for image-assisted maintenance inspection and subsequent manual verification of visible cracks and spalling in slope concrete support structures.

Authors

Institutions

Publication Details

Journal
Journal of Performance of Constructed Facilities
Published
2026-11-14
DOI
https://doi.org/10.1061/jpcfev.cfeng-5677
Primary Topic
Infrastructure Maintenance and Monitoring
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Vision-Based Defect Detection for Maintenance Inspection of Slope Concrete Support Structures

Feng Han, 郭薇 Guo Wei, Yikun Fan, Long Shi et al.
Journal of Performance of Constructed Facilities
Infrastructure Maintenance and Monitoring
article

Vision-Based Defect Detection for Maintenance Inspection of Slope Concrete Support Structures

Feng Han, 郭薇 Guo Wei, Yikun Fan, Long Shi, Yuyao Zhang
article en

Abstract

Abstract Slope concrete support structures are continuously subjected to geological actions, weathering and erosion, and service loads, making their surfaces susceptible to defects such as cracks and spalling. Traditional manual inspection is inefficient, costly, and highly subjective, making it difficult to meet the demands for rapid, large-scale, and standardized inspection of high and steep slopes. To address feature loss and insufficient detection accuracy caused by the complex morphology of cracks and the irregular boundaries of spalling, this study proposes SDIN-YOLO (Space-to-depth convolution, DySample, Inner Complete Intersection over Union, and Normalized Wasserstein Distance-YOLO), a YOLO26-based detection model integrating SPDConv (S), DySample (D), Inner-CIoU (I), and NWD (N) through three optimizations: (1) a SPDConv module is introduced into the backbone network to reduce feature loss during downsampling; (2) the DySample upsampler, which is based on dynamic point sampling, is introduced into the neck network to improve multiscale feature reconstruction; and (3) the Inner-CIoU loss, which combines an auxiliary bounding-box mechanism with CIoU, is adopted and further integrated with the NWD to form a joint loss function, thereby improving bounding-box localization accuracy. The model is trained and evaluated on images annotated with crack and spalling defects, and its detection performance and the effectiveness of each proposed improvement are validated through comparative and ablation experiments. Experimental results showed that SDIN-YOLO achieves a precision of 89.7%, an F1-score of 74.5%, an mean average precision ( mAP 50 ) of 73.5%, and an mAP 50 – 95 of 52.7%, representing improvements of 9.4, 2.0, 1.6, and 5.5 percentage points, respectively, over the baseline YOLO26n model. Experiments under simulated foggy conditions further demonstrated that the model exhibits a certain degree of adaptability to foggy environments. The proposed method can provide defect category and location information for image-assisted maintenance inspection and subsequent manual verification of visible cracks and spalling in slope concrete support structures.

Journal of Performance of Constructed FacilitiesVol. 41(1)
Lanzhou Jiaotong University (CN)
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