462 / 2019-02-28 06:49:59

A Preliminary Study on Visual Features of Urban Street Vitality Using a Convolutional Neural Network

urban street vitality,visual features,convolutional neural networks,Nanjing, China

全体主题 > Artificial Neural Networks

In this study, we tried to identify the features of urban street vitality and evaluate their importance. We selected a certain area in Nanjing, China. Over 433 sites were selected. And we surveyed the area on foot. The investigators were asked to record their evaluation score of the vitality of each site and to take corresponding pictures on site. 370 pictures and recorded score pairs from 243 valid survey sites were then inputted into a neural network to train it, and heat maps were then used to identify the features which lead to high vitality scores. The study found out human, construction site, shop signs and road side and/or walking pavement are key visual features corresponds to the vitality of urban street. The findings can be used to guide the im-provements to the city environment for the better attraction of resi-dents’ activities and communications.