カネミツ　ヒデヒロ   Hidehiro Kanemitsu   金光　永煥    所属   コンピュータサイエンス学部 コンピュータサイエンス学科    職種   専任講師
言語種別	英語
発行・発表の年月	2018/03
形態種別	国際会議論文
査読	査読あり
標題	Effective Feature Selection for Damaged Buildings Using Post-Earthquake Satellite Image with Machine Learning
執筆形態	共著
掲載誌名	Proc. of 2018 RISP International Workshop on Nonlinear Circuits, Communications and Signal Processing (NCSP 2018)
掲載区分	国外
出版社・発行元	RISP
巻・号・頁	pp.315-318
著者・共著者	Jonggeol Park, Masaki Hanada, and Hidehiro Kanemitsu
概要	The final goal of this study is to create data to supportemergency efforts in a disaster affected area by locatingdamaged buildings shortly after the disaster. In this study,prioritizing the practicality of the method for emergencypurposes, we designed a method only to use a singlesatellite image of an affected area, eliminating the use ofcomplex algorithms and auxiliary data. The uniqueness ofour method lies in the application of an object-basedregion segmentation to images and the use of features ofobjects obtained from texture, hierarchical and otherinformation in order to extract damaged buildings. Out of26 features resulting from the analysis of objects, wefound one feature and three combinations of two differentfeatures that are effective in extracting damaged buildings,such as Rectangular fit, Homogeneity, Number of subobjects/Area, and Length of longest of edge/Area.