Hybrid texture analysis of 2D images for detecting asphalt pavement bleeding and raveling using tree-based ensemble methods

Abstract

Since asphalt pavement bleeding and raveling are significant threats to road safety, comfort, and service life, timely and accurate detection of them is desirable and cost-effective to develop pavement maintenance programs. Moreover, due to the fact that both of these distresses (i.e., raveling and bleeding) affect the texture of pavements, simultaneous classification of these distresses can provoke challenges. Thus, this study explores efficient texture analysis of 2D images for bleeding and raveling detection using tree-based ensemble methods. To this end, two scenarios of feature extraction are taken into account using Histogram Equalization (HE), Local Binary Pattern (LBP), and Gray-Level Co-occurrence Matrix (GLCM) techniques. Based on these two feature sets, four tree-based ensemble methods were evaluated for classifying the three classes of Bleeding, Raveling, and No-Bleeding & No-Raveling. The results confirmed that Feature Set B which derived from the integration of HE-GLCM and LBP-GLCM feature extraction provided the most favorable outcome, and gained the high performance of F1-Score ≅ 97 %. Finally, Convolutional Neural Network (CNN) models were constructed using the collected dataset and compared to the proposed method.

Keywords

• Bleeding detection
• Raveling detection
• HE
• LBP
• GLCM
• Tree-based ensemble methods