Loading...
Pavement Crack Assessment Using Satellite Remote Sensing and Deep Learning Model
Al Adab, Maryam
Al Adab, Maryam
Files
35.232-2025.36a Maryam Al Adab COMPRESSED.pdf
Adobe PDF, 1.15 MB
- Embargoed until 2026-09-24
Description
A Master of Science thesis in Civil Engineering by Maryam Al Adab entitled, “Pavement Crack Assessment Using Satellite Remote Sensing and Deep Learning Model”, submitted in July 2025. Thesis advisor is Dr. Tarig Ali. Soft copy is available (Thesis, Completion Certificate, Approval Signatures, and AUS Archives Consent Form).
Abstract
Cracking is one of the common forms of surface distress found in asphalt pavement. It can affect riding quality, structural integrity, and decreases the design life of the pavement. Accurate and timely detection of pavement cracks is crucial for maintenance planning. There are several methods used for assessing pavement condition, and the most common methods include the traditional visual inspection or vehicle-mounted system. However, these methods remain to be time consuming and challenging for large urban networks. Recent advancement is high-resolution satellite imagery and deep learning provide new opportunities for efficient and large-scale pavement condition assessment. This research investigates the feasibility of using high-resolution satellite imagery, combined with deep learning model to detect and classify asphalt pavement crack in the urban context of Los Angeles (LA). LA was selected due to extensive road networks and visible surface distresses pattern that can be captured by satellite. This study uses high-resolution satellite imagery obtained through Google Earth Pro. To achieve this goal, the YOLOv8s-seg model, an advanced variant of the YOLO (You Only Look Once) family, was fine-tuned on a manually labeled crack dataset for realtime object detection and segmentation. The training dataset includes three main types of cracks: alligator cracks, Longitudinal and transverse cracks, and sealed cracks. Images were carefully annotated using the Roboflow platform and augmented to increase data diversity and improve model generalization.The results obtained demonstrate the potential of this approach for automatically detecting and classifying different cracks types from satellite imagery, despite the challenges posed by satellite resolution limits and background noise. The study is among the first to explore pavement crack type classification at a city scale using accessible satellite data and contributes a practical workflow for integrating AI-based surface distress assessment into pavement management strategies. The finding highlights how remote sensing and DL support cost-effective, scalable assessment of asphalt pavement conditions in major cities like Los Angeles.