ニューラルネットワーク分析手法による都市キャノピー幾何情報の推定に関する研究

概要

論文の詳細を見る
The geometric information of cities is indispensable in various areas of urban studies. In this paper, authors report the result of creating the meshed geometry data of Tokyo metropolitan from Detailed Digital Information (10m Grid Land Use), National Population Census, and Establishment and Enterprise Census by simple regression, multiple linear regression analysis, and neural network. The neural network showed finest performance among three methods, its estimating accuracy may reach 0.6 (R^2, coefficient of determination) in all evaluated items in the case that whole targets are used as training signal. To check the general versatility of neural networks, Maharanobis distance was usd to evaluate the similarity of cities, and it was found that neural network is capable of high presition when the Maharanobis distance is under 15.
社団法人日本建築学会の論文
2008-12-30