This study focuses on the pedestrian spaces of campus streets at a university in southern China,utilizing Gradient-weighted Class Activation Mapping(Grad-CAM)to explore pedestrian visual perception experiences.Grad-CAM intuitively highlights key areas in street images that affect pedestrian comfort,automatically identifying and emphasizing visual elements such as vegetation and vehicles,and revealing their impact on the pedestrian experience and visual perception.By comparing Grad-CAM activation maps with eye-tracking data,and incorporating analyses from SHAP and kernel density estimation models,the study summarizes the main street characteristics that shape pedestrian perception,and uncovers the diverse effects of streetscape elements on visual perception.This study provides architects with an analytical method that is easy to understand and apply for the refined design of campus pedestrian spaces.
关键词
校园街道空间/街道步行性/视觉感知体验/深度学习/类激活图/可解释性
Key words
campus street space/street walkability/visual perceptual experience/deep learning/Class Activation Mapping/interpretability
维普
万方数据