Comparison of deep convolutional neural network classifiers and the effect of scale encoding for automated pavement assessment

Deep learning has received a growing interest in recent years for detecting different types of pavement distresses and automating pavement condition assessment. A proper choice of deep learning models is key for successful pavement assessment applications. In this study, the authors first present a comprehensive experimental comparison of state-of-the-art image classification models to evaluate their performances on 11 pavement objects classification. The authors' experiments are conducted in different dimensions of comparison, including deep classifier architecture, effects of network depth, and computational costs. Five convolutional neural network (CNN) classifiers widely used in transportation applications, including VGG16, VGG19, ResNet50, DenseNet121, and a generic CNN (as the control model), are tested with a comprehensive pixel-level annotated dataset for 11 different distress and non-distress classes (UCF-PAVE 2017). In addition, the authors investigate a simple yet effective approach of encoding contextual information with multi-scale input tiles to classify highly random pavement objects in size, shape, intensity, texture, and direction. The authors' comparison results show that the multi-scale approach significantly improves the classification accuracy for all compared deep classifiers at a negligible extra computational cost. Finally, the authors provide recommendations of how to improve the classification performance of deep CNNs for automated pavement condition assessment based on the comparison results.

• Record URL:
  • https://doi.org/10.1016/j.jtte.2022.08.002
• Record URL:
  • http://www.sciencedirect.com/science/article/pii/S2095756423000338
• Availability:
  • Find a library where document is available. Order URL: http://worldcat.org/issn/20957564
• Supplemental Notes:
  • Abstract reprinted with permission of Elsevier.
• Authors:
  • Eslami, Elham
  • Yun, Hae-Bum
  • 0000-0003-0703-8012
• Publication Date: 2023-4

Language

• English

Media Info

• Media Type: Web
• Features: References;
• Pagination: pp 258-275
• Serial:
  • Journal of Traffic and Transportation Engineering (English Edition)
  • Volume: 10
  • Issue Number: 2
  • Publisher: Elsevier
  • ISSN: 2095-7564
  • Serial URL: http://www.journals.elsevier.com/journal-of-traffic-and-transportation-engineering-english-edition
• Publication flags:

  Open Access (libre)

Subject/Index Terms

• TRT Terms: Alternatives analysis; Classification; Evaluation and assessment; Highway maintenance; Machine learning; Neural networks; Pavement distress
• Subject Areas: Highways; Maintenance and Preservation; Pavements;

Filing Info

• Accession Number: 01881294
• Record Type: Publication
• Files: TRIS
• Created Date: Apr 26 2023 9:46AM