Integrated GIS, Remote Sensing, and Machine Learning for Determining Pavement Condition Assessment Rating: A Case Study in Newington, Connecticut

Abstract

The maintenance of transportation systems is vital for road safety and cost reduction. Condition assessment is one key aspect of this process. However, the use of computer vision and machine learning (ML) has the potential to streamline and improve this process. This study investigates the integration of geographic information systems (GIS), aerial imagery, and ML to enhance pavement maintenance requirements by developing a pavement condition assessment rating (PCAR) based on remote sensing, severity level, and crack percentages. The objective of this paper is to identify roads with and without cracks to determine the PCAR for effective planning and resource allocation. A systematic workflow included data collection, preprocessing, support vector machines (SVM) classification, accuracy assessment, and visualization. Twelve satellite images (0.25 ft resolution) from the USGS database covered part of Newington, Connecticut. The reclassified model achieved an overall accuracy of 72% with a substantial correlation (kappa coefficient of 67%). The roads with cracks had a producer accuracy of 80%, indicating that the algorithm successfully identified and included 80% of the reference pixels belonging to this class. After this, the PCAR was obtained, e.g., Cambria Avenue had a cracking area percentage of 26.72%, with a PCAR classified as “Fair”, while roads like Adrian Avenue showed lower cracking area percentages, indicating PCAR of “Good”. The results highlight the effectiveness of ML and aerial imagery for classifying roads and obtaining PCAR. These findings assist in prioritizing road maintenance decisions while converting crack percentages into severity indices and aid in assessing pavement conditions and planning rehabilitation efforts.