Visualization for Explanation of Deep Learning-Based Defect Detection Model Using Class Activation Map

Recently,convolutional neural network(CNN)-based visual inspec-tion has been developed to detect defects on building surfaces automatically.The CNN model demonstrates remarkable accuracy in image data analysis;however,the predicted results have uncertainty in providing accurate informa-tion to users because of the“black box”problem in the deep learning model.Therefore,this study proposes a visual explanation method to overcome the uncertainty limitation of CNN-based defect identification.The visual repre-sentative gradient-weights class activation mapping(Grad-CAM)method is adopted to provide visually explainable information.A visualizing evaluation index is proposed to quantitatively analyze visual representations;this index reflects a rough estimate of the concordance rate between the visualized heat map and intended defects.In addition,an ablation study,adopting three-branch combinations with the VGG16,is implemented to identify perfor-mance variations by visualizing predicted results.Experiments reveal that the proposed model,combined with hybrid pooling,batch normalization,and multi-attention modules,achieves the best performance with an accuracy of 97.77%,corresponding to an improvement of 2.49%compared with the baseline model.Consequently,this study demonstrates that reliable results from an automatic defect classification model can be provided to an inspector through the visual representation of the predicted results using CNN models.

A GIS-Based Framework for Real-Time Debris-Flow Hazard Assessment for Expressways in Korea

Debris flows caused by heavy rainfall in mountain areas near expressways lead to severe social and economic losses and sometimes result in casualties.Therefore, the development of a real-time system for debris-flow hazard assessment is necessary to provide preliminary information for rapid decision making about evacuations or restoration measures, as well as to prevent secondary disasters caused by debris flows. Recently,various map-based approaches have been proposed using multi-attribute criteria and assessment methods for debrisflow susceptibilities. For the macrozonation of debris-flow hazard at a national scale, a simplified method such as the Korea Expressway Corporation(KEC) debris-flow hazard assessment method can be applied for systematic analysis based on geographic information systems(GIS) and monitoring networks. In this study, a GIS-based framework of real-time debris-flow hazard assessment for expressway sections is proposed based on the KEC debris-flow hazard assessment method. First, the KEC-based method was standardized in a systematic fashion using Arc GIS,enabling the objective and quantitative acquisition of various attribute datasets. The quantification of rainfall criteria also was considered. A safety management system for debris-flow hazard was developed based on the GIS platform. Finally, the method was applied and verified on three expressway sections in Korea. The grading standard for each individual influencing attribute was subsequently modified to more accurately assess the debris-flow hazards.

PotholeEye^(+): Deep-Learning Based Pavement Distress Detection System toward Smart Maintenance

We propose a mobile system,called PotholeEye+,for automatically monitoring the surface of a roadway and detecting the pavement distress in real-time through analysis of a video.PotholeEye+pre-processes the images,extracts features,and classifies the distress into a variety of types,while the road manager is driving.Every day for a year,we have tested PotholeEye+on real highway involving real settings,a camera,a mini computer,a GPS receiver,and so on.Consequently,PotholeEye+detected the pavement distress with accuracy of 92%,precision of 87%and recall 74%averagely during driving at an average speed of 110 km/h on a real highway.

Evaluation of early-age strains and stresses in roller-compacted concrete pavement

In the structural design of a roller-compacted concrete pavement(RCCP), it is crucial to estimate strain and stress developments in the RCCP slab realistically. Since the RCCP mix uses less cement and lower amount of water, shrinkage strain and concrete temperature during the hardening stage are expected to be reduced as compared to those of conventional concrete mixture, resulting in a reduction of the concrete early-age deformation and stress developments in the RCCP slab. In this paper, early-age concrete strain and stress developments in RCCP slab subjected to environmental loads were evaluated. A full-scale test section of RCCP under real climatic conditions was monitored. The early-age total strains,stress-independent strains, shrinkage strains, and coefficient of thermal expansion(CTE) of the RCCP were measured and analyzed. Using the results of measured strains, in-situ CTE and shrinkage strain, and temperature, the early-age concrete stress development is computed by incorporating a viscoelastic property of the early-age concrete. The results revealed that the shrinkage strain of the RCCP is quite low as compared to that of conventional concrete. The early-age stress developments in the RCCP slab are strongly governed by the thermal-induced stresses. Shrinkage-induced stresses were quite small and might be negligible in a preliminary estimation of early-age stress developments in the RCCP slab.