A high precision crack classification system using multi-layered image processing and deep belief learning

Road surfaces experience fatigue stress and loading, which often lead to cracks on the surface. The cracks might cause serious damage, and therefore, early detection can reduce the road maintenance cost. Traditional inspection methods are carried out by humans and are slow, costly and hazardous. To improve accuracy and reduce the hazards of current crack detection methods, this paper proposes a new autonomous crack detection system (ACDS) that can be used in any autonomous vehicles (UAVs). ACDS consists of three stages: image acquisition, image processing, and classification. The image processing stage consists of five parallel filtering methods, which remove noise and extract features from the images. In the classification stage, five deep belief network (DBN) classifiers separately analyse the images to detect cracks. The dataset used in this paper contains 15,000 RGB and infrared images, with or without cracks. The results show the high precision of the proposed system.

• Record URL:
  https://doi.org/10.1080/15732479.2019.1655068
• Availability:
  • Find a library where document is available. Order URL: http://worldcat.org/issn/15732479
• Supplemental Notes:
  • © 2019 Informa UK Limited, trading as Taylor & Francis Group. Abstract republished with permission of Taylor & Francis.
• Authors:
  • Jo, Jun
  • Jadidi, Zahra
• Publication Date: 2020-2

Language

• English

Media Info

• Media Type: Web
• Features: Figures; References;
• Pagination: pp 297-305
• Serial:
  • Structure and Infrastructure Engineering
  • Volume: 16
  • Issue Number: 2
  • Publisher: Taylor & Francis
  • ISSN: 1573-2479
  • EISSN: 1744-8980
  • Serial URL: http://www.tandfonline.com/loi/nsie20

Subject/Index Terms

• TRT Terms: Automatic data collection systems; Autonomous vehicles; Classification; Detection and identification systems; Image processing; Machine learning; Pavement cracking
• Subject Areas: Data and Information Technology; Highways; Maintenance and Preservation;

Filing Info

• Accession Number: 01730610
• Record Type: Publication
• Files: TRIS
• Created Date: Feb 7 2020 9:39AM