Development of a Framework for Identifying Asphalt Pavement Cracking Distresses Using Machine Learning [Research Brief]

Details

• Creators:
  Cheng, DingXin
• Corporate Creators:
  San Jose State University. College of Business. Mineta Transportation Institute ; California State University, Chico. California Pavement Preservation Center
• Corporate Contributors:
  United States. Department of Transportation. University Transportation Centers (UTC) Program ; United States. Department of Transportation. Office of the Assistant Secretary for Research and Technology
• Subject/TRT Terms:
  Asphalt Mixtures Asphalt Pavements Computer Vision Costs Detectors Deterioration Machine Learning Maintenance Pavement Condition Pavement Cracking Pavement Distress

  More +
• Publication/ Report Number:
  2435
• Resource Type:
  Brief
• Geographical Coverage:
  United States
• Edition:
  Research Brief
• Corporate Publisher:
  San Jose State University. College of Business. Mineta Transportation Institute
• Abstract:
  The research team developed and tested a prototype machine learning (ML) framework to automate the identification of asphalt pavement cracking. The framework combines deep learning and image processing, trained using a large dataset of annotated pavement images collected from California roadways. Each image was labeled according to standardized distress types—based on ASTM, FHWA, Caltrans, and MTC manuals—to ensure consistency across different severity levels and distress categories. A convolutional neural network (VGG16) was used for image classification—essentially teaching the system The research team developed and tested a prototype machine learning (ML) framework to automate the identification of asphalt pavement cracking. The framework combines deep learning and image processing, trained using a large dataset of annotated pavement images collected from California roadways. Each image was labeled according to standardized distress types—based on ASTM, FHWA, Caltrans, and MTC manuals—to ensure consistency across different severity levels and distress categories. A convolutional neural network (VGG16) was used for image classification—essentially teaching the system to recognize what type of crack appears in each image—while a YOLO model performed real-time object detection and localization, identifying where cracks appear on the pavement and outlining them in the image. The models were trained and validated using a 70/20/10 data split, with data augmentation techniques applied to improve robustness under varying lighting and texture conditions, so the system can still perform well even when images are taken at different times of day or on different pavement surfaces.
• Format:
  PDF
• Collection(s):
  University Transportation Centers
• Main Document Checksum:
  urn:sha-512:7ffde83d7b120519412ace649ec6915953f1f80dcd7dea4bca1f2a0f2c2d8e99db39a2f4de86dd489b241ef8e20afe305c010ae6990fd71b4733bdffab36772e
• Download URL:
  https://rosap.ntl.bts.gov/view/dot/89509/dot_89509_DS1.pdf
• File Type:
  [PDF - 220.05 KB ]