Revealing emotional responses to urban environmental elements through street view data and deep learning

Environmental characteristics affect how individuals perceive an environment, and the emotions created by environmental characteristics originate from subjective feelings. Despite cities being crucial human living spaces, few studies have used geospatial technology to determine the relationship between urban environments and emotions. Therefore, this study explored the effects of urban environmental characteristics on emotions by surveying 50 sampling areas in Taipei City. Deep learning was performed with the DeepLab V3 architecture in combination with the LaDeco tool to identify environmental characteristics in over 200,000 Google Street View (GSV) images. These characteristics were divided into five major types, namely, vegetationscapes, waterscapes, streetscapes, landformscapes, and archiscapes, then further classified into 53 categories. To identify the emotions related to urban environments, 2090 participants who were asked to view GSV videos and report their emotions. Subsequent multiple regression analyses revealed that in vegetationscapes and waterscapes, grass and fountains induced positive emotions, whereas trees reduced negative emotions. Meanwhile, dense, old, and disorganized buildings, such as hovels, reduced positive emotions. The results of this study may serve as a reference to help designers create an urban environment that fosters positive emotions.