Artificial Intelligence and Data Science
Online ISSN : 2435-9262
Deterioration classification of tunnel cable rack based on color and texture features extracted from drive-by images
Takahiro TSUBOTA, Takuto KONDO
Author information
JOURNAL OPEN ACCESS

2026 Volume 7 Issue 3 Pages 533-541

Details
Abstract

This study investigates the feasibility of classifying the deterioration levels of tunnel equipment using color and texture features extracted from drive-by images. A total of 1,168 images of cable-rack cross members were obtained from four survey sections in two expressway tunnels. Eleven image features were calculated based on the hue and saturation components of the HSV color space and gray-level co-occurrence matrices. The images in each survey section were arranged in spatial order and divided into spatially continuous training and test subsets. Three machine-learning models, LightGBM, XGBoost, and Random Forest, were then developed to classify deterioration into three levels, A, B, and C. Among the models, LightGBM achieved the highest classification accuracy of 89.20%. Its recall values for Classes A and C were 97.04% and 94.07%, respectively, whereas the recall for the intermediate Class B was 68.29%. SHapley Additive exPlanations were further applied to interpret the LightGBM model. The mode of saturation and the proportion of red pixels were identified as the most influential features for distinguishing Classes A and C, while the contributions of the texture features were relatively limited. These results indicate that global color-distribution features extracted from drive-by images can support the preliminary screening of tunnel equipment and help prioritize facilities requiring detailed inspection.

Content from these authors
© 2026 Japan Society of Civil Engineers
Previous article Next article
feedback
Top