A novel method of urban landscape perception based on biological vision process

Abstract

Urban landscape perception is essential for understanding the interaction between individuals and the built environment, impacting urban space quality improvement. This study bridges the gap in comprehending the mechanisms, processes, and content of landscape perception that previous studies have not fully addressed. By integrating urban landscape studies with the biological vision process, a new theoretical framework is proposed, which includes an index system with 4 dimensions: color features, landscape elements, spatial forms, and landscape imagery, consisting of 30 indicators. Furthermore, a novel method leveraging Large Vision Models for color analysis, semantic segmentation, object detection, and depth prediction is introduced. This method allows for the accurate extraction of objective features of urban landscapes and uses the Random Forest to analyze the nonlinear relationships between objective features and subjective perceptions. An empirical study conducted in Chongqing demonstrates that color features and spatial forms significantly influence landscape perception, similar to the landscape elements. Moreover, ablation experiments demonstrate that our approach, based on the biological vision process, improves accuracy and fit compared to existing methods. This study elucidates crucial factors affecting landscape perception, refines and generalizes perception methods, and aids planners in navigating complex scenarios, contributing to the practical application and widespread adoption of landscape perception in urban planning.