Automatic damage detection using anchor-free method and unmanned surface vessel

Damage detection is a fundamental measure to study the security status of infrastructure. But some key parts are so difficult to enter by manpower and equipment, making it extremely challenging to detect (e.g. the bottoms of small and medium bridges, urban underground culverts, etc.). Therefore, this paper proposes a systematic solution including two parts: a deep learning method and an unmanned surface vessel (USV) to efficaciously detect damage. For the detection algorithm, a novel anchor-free network, CenWholeNet, which focuses on center points and holistic information, is proposed. The attention mechanism was also introduced into the model innovatively and a parallel attention module (PAM) was put forward. Sufficient ablation experiments substantiate that PAM can enhance the representation capability of models with negligible computational overheads. For the smart device, a USV system without the global positioning systems (GPS) navigation is proposed, supporting real-time transmission of lidar and video information. The presented solution was applied to the inspection of a bridge group, which verified the effectiveness and applicability of the solution to complex projects. Furthermore, the results corroborate that, compared with the influential objective detection methods, such as Faster R-CNN and YOLOv5, CenWholeNet is more suitable for the detection of multiple diseases with variable slenderness ratios and complex shapes. Therefore, CenWholeNet is expected to become a new paradigm of intelligent damage detection for infrastructure.

• Record URL:
  https://doi.org/10.1016/j.autcon.2021.104017
• Record URL:
  http://www.sciencedirect.com/science/article/pii/S0926580521004684
• Availability:
  • Find a library where document is available. Order URL: http://worldcat.org/issn/09265805
• Supplemental Notes:
  • © 2021 Published by Elsevier B.V. Abstract reprinted with permission of Elsevier.
• Authors:
  • He, Zhili
  • Jiang, Shang
  • Zhang, Jian
  • Wu, Gang
• Publication Date: 2022-1

Language

• English

Media Info

• Media Type: Web
• Features: References;
• Serial:
  • Automation in Construction
  • Volume: 133
  • Issue Number: 0
  • Publisher: Elsevier
  • ISSN: 0926-5805
  • Serial URL: http://www.sciencedirect.com/science/journal/09265805

Subject/Index Terms

• TRT Terms: Autonomous vehicles; Detection and identification; Infrastructure; Loss and damage; Ships
• Subject Areas: Marine Transportation; Vehicles and Equipment;

Filing Info

• Accession Number: 01785795
• Record Type: Publication
• Files: TRIS
• Created Date: Oct 26 2021 11:20AM