PCI Estimation Method for Evaluating Asphalt Deterioration Leveraging SAM and YOLOv8

This study presents a deep-learning-based approach to estimating the pavement condition index (PCI) to evaluate asphalt deterioration. Current machine learning and deep learning models are integrated to optimize traditional methods, which are expensive and time consuming. Types of asphalt deterioration have been determined and classified using the “You Only Look Once” version 8 (YOLOv8) model. The model’s high-speed detection capabilities were demonstrated by its high accuracy, with mAP@50 achieving 99.4% and mAP@50–95 at 88.9%. The segment anything model (SAM), developed by Meta, was used to segment and identify the area of deterioration. However, because SAM requires object localization to perform segmentation, YOLOv8-generated bounding boxes served as the foundation for this process. The data set used in this study comprises seven distinct classes representing various types of asphalt deterioration. For detection and segmentation tasks, the YOLOv8 model was trained for 200 epochs. Metrics such as precision, recall, and F1 scores were used to evaluate performance. Notably, the YOLOv8 model outperformed other models in segmentation tests for crack class by concentrating on the exact deterioration areas and reducing background noise. Based on the locations of various deterioration types, PCI calculations were carried out, and the findings were exported to a comma-separated values (CSV) file for further analysis. The findings demonstrate that an interpretable and effective method for automated asphalt deterioration recognition and PCI estimation is offered by the combination of the YOLOv8 and SAM models. This approach offers enhanced decision-support mechanisms for intelligent transportation systems and road management, contributing to the development of smarter and more sustainable infrastructure solutions.