[4Rin1] Interactive 2

[4Rin1-32] Extraction of Story-collapsed Buildings Based on Image Analysis

〇Ken Kawabe¹, Kei Horie¹, Munenari Inoguchi², Masashi Matsuoka³ (1.MS&AD InterRisk Research & Consulting, Inc., 2.University of Toyama, 3.Tokyo Institute of Technology)

Keywords:Deep Learning, Story-collapse, Photographs of Damaged Buildings, Earthquake

In the 1995 Kobe Earthquake, many lives were lost due to the collapse of buildings called "story-collapse" without survival space. In order to reduce human casualties, it is important to identify buildings at high risk of story-collapse and to promote countermeasures for earthquake disaster. In this study, we propose the method for extraction of story-collapsed buildings based on image analysis using deep learning models. Using the data set of photographic images of damaged buildings taken during damage investigation by the local government in the 2016 Kumamoto Earthquake, we developed classification models using convolutional neural network and extracted story-collapsed buildings.