Integrating Machine Learning,Remote Sensing,and a Geographic Information System to Determine Road Surface Type and Construction Year from High-Resolution and Multispectral Aerial Imagery: A Case Study in Ames,Iowa

Abstract

Road pavement data plays a crucial role in establishing an effective pavement management system and enables agencies to make more accurate decisions on maintenance and rehabilitation plans. Traditional pavement data collection methods are time-consuming, costly, and labor-intensive, limiting their scalability for large-scale infrastructure assessment. Utilizing satellite/aerial imagery for extracting road pavement data is becoming increasingly popular because of the availability of high-resolution images, advancements in image processing and remote sensing techniques, enhanced computational power, and the presence of geographic information systems (GISs). This novel study integrates machine learning (ML), remote sensing, and a GIS to develop an automated tool based on a robust deep-learning model to extract the construction year from time-series aerial imagery. This study also employs pixel classification to automate differentiation between asphalt concrete (AC) and Portland cement concrete (PCC) road surface pixels using multispectral aerial imagery. The study area is the city of Ames in Story County, Iowa, where the model’s performance is rigorously assessed. Results demonstrate high accuracy in predicting road construction years, achieving 94% for both recall and precision. The AC and PCC surfaces were classified with 89% overall accuracy, 95.38% and 77.14% recall rates, respectively, and PCC demonstrated a high precision of 90%, compared to AC’s 88.57%. This study underscores the need for high-resolution multispectral imagery (30 cm or less) captured in spring with cloud cover below 10% for optimal accuracy. These findings highlight the potential of remote sensing and ML in improving pavement data collection, enhancing accuracy, and reducing manual effort. Higher-resolution imagery with an extended spectrum can further improve accuracy.