Research on visual perception evaluation of urban riverside greenway landscape based on deep learning

  Objective  The visual quality of urban waterfront landscape has an important impact on the quality of urban waterfront landscape and urban space. With the rapid development of urbanization, it is the development trend of waterfront landscape in the future to screen the factors, which have important impacts on the visual perception of urban waterfront greenway, and realize the intelligent analysis of waterfront landscape through the quantitative processing of visual perception.

  Method  Based on deep learning algorithm, this paper simulates human visual perception and trains a set of image semantic segmentation model for urban riverside greenway landscape. And the multiple linear regression model was used to explore the relationship between landscape features and visual perception. Taking the second ring water system in Beijing as an example, this paper analyzes the landscape characteristics and promotion strategies under the visual perception.

  Result  (1) The trained semantic segmentation model of urban riverside greenway image achieved an accuracy of 0.93. (2) The results of regression model showed that in 10 indicators (green visual index (GVI), blue visual index (BVI), hardening reventment index (HRI), riverside building density (RBD), bridge visibility (BV), interference factor index (IFI), waterfront openness (WO), waterfront guardrail (WG), road width index (RWI) and ratio of trees to grass (RTG)) system, five items had significant effects on visual perception, i.e. GVI, WO, BVI, IFI and RTG, of which GVI and WO were positively correlated with visual perception, and the other three items were negatively correlated with visual perception. (3) GVI of the second ring river system in Beijing was higher in the north than in the south, WO and BVI were relatively homogeneous, IFI was generally low, and RTG showed the characteristics of division by river channel. (4) When improving the visual perception effect, we can focus on the five indicators with significant impact and weigh according to the strength of the impact.

  Conclusion  This paper provides a measurement method based on image semantic segmentation for the study of urban waterfront landscape, and provides more possibilities for the quantitative analysis of large-scale waterfront greenway landscape from the humanistic perspective.