Post-Hurricanes Roadway Closure Detection using Satellite Imagery and Semi-Supervised Ensemble Learning

After hurricanes, roadway damage assessment is critical to emergency responders and city authorities. In this paper, the authors propose an automated semi-supervised approach to identify tree debris along roadways to improve the efficiency of damage assessment and allow faster debris cleaning operations. The solution uses two high-resolution satellite images taken before and after a hurricane. The proposed methodology is an ensemble learning machine using an unsupervised autoencoder-based feature extractor, an unsupervised vegetation coverage estimator, and a weakly-supervised tree segmentation. The authors show that such a combination can increase precision and accuracy. The authors' solution has been tested with a case study on Hurricane Michael, which hit Tallahassee, the capital of Florida, in October 2018.

• Record URL:
  https://annualmeeting.mytrb.org/OnlineProgram/Details/15707
• Supplemental Notes:
  • This paper was sponsored by TRB committee AMR20 Standing Committee on Disaster Response, Emergency Evacuations, and Business Continuity.
• Corporate Authors:

  Transportation Research Board

  ,    
• Authors:
  • Gazzea, Michele
  • 0000-0003-0759-4887
  • Karaer, Alican
  • 0000-0002-6704-0379
  • Balafkan, Nozhan
  • Ozguven, Eren Erman
  • 0000-0001-6006-7635
  • Arghandeh, Reza
  • 0000-0002-0691-5426
• Conference:
  • Transportation Research Board 100th Annual Meeting
  • Location: Washington DC, United States
  • Date: 2021-1-5 to 2021-1-29
• Date: 2021

Language

• English

Media Info

• Media Type: Digital/other
• Features: Figures; Maps; Photos; References;
• Pagination: 18p

Subject/Index Terms

• TRT Terms: Highway maintenance; Hurricanes; Image analysis; Loss and damage; Machine learning; Remote sensing
• Identifier Terms: Hurricane Michael, 2018
• Geographic Terms: Tallahassee (Florida)
• Subject Areas: Highways; Maintenance and Preservation; Security and Emergencies;

Filing Info

• Accession Number: 01763675
• Record Type: Publication
• Report/Paper Numbers: TRBAM-21-00892
• Files: TRIS, TRB, ATRI
• Created Date: Feb 4 2021 10:57AM