SAFETYLIT WEEKLY UPDATE

We compile citations and summaries of about 400 new articles every week.
RSS Feed

HELP: Tutorials | FAQ
CONTACT US: Contact info

Search Results

Journal Article

Citation

Sakamoto J, Nakamura J. Proc. Conf. Japan Soc. Traffic Eng. 2022; 42: 361-365.

Vernacular Title

深層学習を援用した洪水時における道路浸水状況の自動検出手法の提案

Copyright

(Copyright © 2022, Japan Society of Traffic Engineers)

DOI

10.14954/jsteproceeding.42.0_361

PMID

unavailable

Abstract

When a large-scale flood occurs, it is extremely important to quickly identify the inundation area in formulating an emergency plan. This study proposes a method for automatically detecting flooded road sections. Using an object detection algorithm, a learning model developed using aerial photographs taken during past floods is applied to aerial photographs taken during other floods to verify the suitability of inundation conditions for road sections. do. In the verification, in order to detect in detail how much road is flooded, GIS is used to divide the aerial photograph into meshes, and the road links are superimposed and visualized for each mesh. As a result, the precision of the learning model was 81% to 91%, and the inundation recall was 84% ​​to 94%. In addition, there were many misjudgments in images where it was difficult to judge the flood situation, and in villages in mountainous areas.

===

大規模な洪水が発生した際においては,応急対策計画を立案する上で,迅速な浸水範囲の特定が極めて重要となる.本研究は,洪水により浸水した道路区間を自動的に検出する手法を提案するものである.物体検知アルゴリズムを用いて,過去に発生した洪水時に撮影された航空写真を用いて開発した学習モデルを,別の洪水時に撮影された航空写真に適用し,道路区間の浸水状況の適合状況について検証する.検証にあたっては,どの程度の道路が浸水しているのかを詳細に検出するため,GIS を援用して航空写真をメッシュに分割し,メッシュごとに道路リンクを重ね合わせて視覚化する.その結果,学習モデルの適合率は 81%~91%,浸水再現率は 84%~94%という極めて精度の高いものとなった.また,浸水状況を判断しづらい画像や,山間部の集落などでは誤判定が多くみられた.


Language: ja

Keywords

GIS; YOLO; 浸水区間; 深層学習

NEW SEARCH


All SafetyLit records are available for automatic download to Zotero & Mendeley
Print