Development of damage classification rules suitable for machine learning and a method to automatically assess the degree of building damages

2023 Fiscal Year Final Research Report

• Project/Area Number: 21H01579
• Research Category: Grant-in-Aid for Scientific Research (B)
• Allocation Type: Single-year Grants
• Section: 一般
• Review Section: Basic Section 25020:Safety engineering-related
• Research Institution: Tokoha University
• Principal Investigator: TANAKA Satoshi 常葉大学, 大学院・環境防災研究科, 教授 (90273523)
• Co-Investigator(Kenkyū-buntansha): 重川 希志依 常葉大学, 社会環境学部, 名誉教授 (10329576) ; 松岡 昌志 東京工業大学, 環境・社会理工学院, 教授 (80242311) ; 鱒沢 曜 明星大学, 建築学部, 准教授 (90533141)
• Project Period (FY): 2021-04-01 – 2024-03-31
• Keywords: 深層学習 / 建物被害 / 画像検出 / インスタンス・セグメンテーション / 建物被害判定 / スマート・インスペクション / 画像分類

Outline of Final Research Achievements

In this study, we developed a core technology to automate the building damage assessment by introducing deep learning technology, which until now has relied on the experience of inspectors. Specifically, using a building damage assessment for damage certification as an example, we verified learning models under various conditions and constructed an optimal model to replace the inspector's visual assessment of building damage with a photographic assessment using deep learning. We also developed a smartphone application that incorporates the completed model, and examined its effectiveness and usability through a demonstration experiment for municipal employees. As a result of the verification, it was confirmed that both the inspection time and evaluation accuracy were greatly improved compared to the conventional method.

Free Research Field

地震防災

Academic Significance and Societal Importance of the Research Achievements

本研究では、これまで調査員の技量や経験に依存するため、評価のばらつきや迅速性に問題があった建物被害調査の実務に深層学習技術を適用し、人間による調査が機械に代替可能であることをあきらかにした。この仕組みを実用化すれば、建物被害調査における人間の作業は、損傷の発見、損傷場所の記録、損傷箇所の写真撮影の３つまで減らすことが可能になり、被災者という未開発の資源の活用や、災害時の希少資源である専門技術者の有効活用にもつながり、災害対応上その意義はきわめて大きい。