Superpixel-Based Image Analysis for Stripping Detection in Asphalt Mixtures with ANN and U-Net Models
Journal of Materials and Mechatronics: A · 26.12.2025 · 6 (2) · 442-462
DOI: 10.55546/jmm.1769352 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Materials and Mechatronics: A, online erschienen am 26.12.2025. Verfasst von Kadir Akgöl und 1 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 23 Literaturangaben.
Abstract (Originalsprache)
The reliable detection of stripping in asphalt mixtures is a critical challenge for pavement performance evaluation, as conventional physical tests rely heavily on subjective observation and lack reproducibility. This study proposes an image-based quantitative method that integrates geometric standardization, superpixel segmentation, and feature extraction to enhance the objectivity of stripping assessment. Petri dish images were first standardized through square cropping and bicubic resampling to ensure comparability across samples. Superpixels were then generated, and multiple spatial, geometric, photometric, and texture-based features were extracted, including distance-to-center, compactness, local color similarity, and global color deviation. Automatic background labeling was achieved through a color-based masking approach validated by visual inspection. The extracted feature set was subsequently employed for supervised classification using artificial neural networks (ANNs), with model performance evaluated against reference segmentations. The results demonstrate that the proposed method achieves high classification accuracy, with robust generalization across multiple sample sets. In particular, ANN-based predictions exhibited superior discriminative capability in distinguishing stripped from coated aggregate regions, outperforming U-Net segmentation under identical input conditions. The findings highlight that incorporating contextual descriptors, such as black pixel ratio and blue-background masking, significantly improves classification robustness in low-contrast and noisy regions. Overall, the proposed framework provides a reproducible and efficient alternative to conventional stripping tests, enabling reliable quantitative evaluation of asphalt mixture performance. This study contributes to the advancement of automated image analysis methods in pavement engineering and establishes a foundation for broader integration of computer vision into asphalt durability assessment.
Zitieren
APA
Akgöl, K., & Tuna, M. C. (2025). Superpixel-Based Image Analysis for Stripping Detection in Asphalt Mixtures with ANN and U-Net Models. Journal of Materials and Mechatronics: A, 6(2), 442-462. https://doi.org/10.55546/jmm.1769352
Vancouver
Akgöl K, Tuna MC. Superpixel-Based Image Analysis for Stripping Detection in Asphalt Mixtures with ANN and U-Net Models. Journal of Materials and Mechatronics: A. 2025;6(2):442-462. doi:10.55546/jmm.1769352
BibTeX
@article{Akgl2025,
author = {Akgöl, Kadir and Tuna, Mehmet Can},
title = {Superpixel-Based Image Analysis for Stripping Detection in Asphalt Mixtures with ANN and U-Net Models},
journal = {Journal of Materials and Mechatronics: A},
year = {2025},
volume = {6},
number = {2},
pages = {442--462},
doi = {10.55546/jmm.1769352}
}
EUVIMED