Speaker
Description
Street perception encompassed multiple layers, from basic visual elements to complex cognitive interpretation. While traditional urban perception studies primarily focused on visual features, they often overlooked the deeper, multidimensional aspects of perception. This research introduced a novel unsupervised multi-feature integration framework for analyzing streetscape change perception, combining multimodal large language models with deep learning techniques. The framework enabled intelligent analysis of Shanghai's streetscape evolution patterns through an automated process of training data generation, model development, and model application. Our results demonstrated that ChatGPT's evaluation of streetscape transformation extended beyond simple visual changes to encompass complex urban dynamics. The model exhibited sophisticated analytical capabilities by incorporating both socioeconomic factors, such as commercial vitality, and physical transformations, including construction and renovation activities, in assessing environmental quality changes. The study made significant methodological contributions by establishing an intelligent, reproducible framework for dynamic streetscape evaluation. Furthermore, it uncovered fundamental patterns in metropolitan streetscape evolution, providing valuable insights for informed urban renewal policy-making.
| Keywords | Streetscape Change; Perception; ChatGPT; Deep Learning |
|---|---|
| Best Congress Paper Award | Yes |
