Publication: Image-based texture analysis and machine learning modeling of asphalt pavement anti-skidding performance.
Loading...
Date
2025-12
Authors
Zhao, Yu
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The anti-skidding performance of asphalt pavement not only significantly affects road safety, but also affects the comfort and durability of the pavement. During the completion acceptance stage of asphalt pavement, it is necessary to conduct reasonable testing and evaluation of the anti-skidding performance of the pavement. In response to the limitations of traditional asphalt pavement anti-skidding performance evaluation methods, such as reliance on specialized equipment, susceptibility to human factors, and inefficiency in large-scale applications, this study proposes a comprehensive evaluation method that integrates advanced image processing and machine learning algorithms. In this study, a texture image acquisition device was designed to standardize the collection of pavement texture information. Additionally, after cropping, denoising and grayscale processing, statistical features are extracted via Gray-Level Co-occurrence Matrix, while fractal features are derived using fractal theory. Correlation analysis and clustering screening mechanisms are combined to optimize feature combination. Energy, Entropy, Fractal Dimension and Multifractal Spectral Width are selected to form a multi-scale characterization system for asphalt pavement texture features. During the modelling phase, large sample databases from Anhui and Hubei provinces involving different pavement surfacing types (AC20, AC13, and SMA13) were established to compare the optimization strategies and adaptability of four models, namely Multiple Linear Regression, Random Forest, XGBoost, and BP Neural Network models. The results demonstrate demonstrated that the XGBoost model achieved superior accuracy, with an R²of 0.92 and RMSE of 0.04, and its parallel computing capability ensures superior efficiency. Comparison between the developed XGBoost model and the field assessments on pavement anti-skidding performance, both have shown high performance correlation despite different gradation types were adopted, making it the optimal model for evaluating anti-skidding performance based on asphalt pavement texture features. This study has successfully established an evaluation method that
integrates machine vision and machine learning technology, breaking through the limitations of single texture feature representation and realizing the application of machine learning in the evaluation of anti-skidding performance of asphalt pavement, providing reliable technical support for road maintenance decision-making.