An Adaptive Aggregate-Mortar Segmentation Algorithm with Boundary Refinement and Adhesion Separation for Asphalt Mixture Mesostructural Characterization

Accurate and automated mesostructural characterization of asphalt mixtures remains challenging because image segmentation errors can compromise the extraction of aggregate boundaries and mortar-related features. To address this challenge, this study developed an automated image-based approach for mesostructural characterization and investigated the relationship between mesostructural characteristics and cracking resistance. The proposed approach integrates G-channel enhancement, Otsu-based adaptive thresholding, and roundness-guided boundary refinement to achieve automated aggregate segmentation and adhesion separation. Mortar film thickness distribution and spatial uniformity were quantified from the processed images and related to cracking resistance through partial least squares regression (PLSR). The results indicate that the integration of G-channel enhancement and Otsu-based optimization improved aggregate-mortar separability and enabled automatic determination of the optimal enhancement level and segmentation threshold. Roundness-based boundary refinement and adhesion separation achieved high segmentation accuracy, with pixel accuracy above 93%, mIoU above 86%, and F1 scores above 91%. Decreasing asphalt content shifted the mortar film thickness distribution toward thinner films, while spatial uniformity remained relatively insensitive. RAP incorporation further reduced mortar film thickness and deteriorated spatial uniformity, indicating increased mesostructural heterogeneity. Moreover, asphalt content and mortar film thickness were found as the dominant factors governing cracking resistance in virgin mixtures, whereas spatial uniformity indicators played a more important role in RAP mixtures. These mesostructural parameters show strong potential for application in mixture design and durable pavement engineering.