Proposal of an Automatic Road Detection Method in a flood event Using Deep Learning / 深層学習を援用した洪水発生後の道路浸水状況自動判別手法の提案
- Resource Type
- Journal Article
- Authors
- Jun SAKAMOTO; Junya NAKAMURA; 中村 純也; 坂本 淳
- Source
- 交通工学論文集 / JSTE Journal of Traffic Engineering. 2023, 9(4)8
- Subject
- Aerial photograph
Deep learning
GIS
YOLO
flooded section
浸水区間
深層学習
航空写真
- Language
- Japanese
- ISSN
- 2187-2929
Identifying an inundation area after a flood event is essential to developing an emergency response plan. This study proposes a method to automatically determine disrupted road segments by floods using image recognition technology, a deep learning method. The algorithm used in this study is YOLO (You only look once) v3, developed in 2018. A model developed using aerial photographs taken during a past flood event is applied to aerial photographs taken during another flood event to verify the conformity of inundation conditions of road segments. The model visualizes the inundation status of roads by determining the presence or absence of inundation on a 100-meter mesh basis using aerial photographs and integrating the information on whether the mesh includes road links. The results showed that the model developed in this study could discriminate road flooding with more than 80% accuracy. However, there are frequent misclassified images where it was difficult to determine the inundation status and images that included fields.