Quantitative Characterization of Highway Landscape Space Visual Perception Based on Deep Learning

With the advent of the modern transportation era, road design needs to emphasize landscape and recreational functions more than traditional safety functions. However, quantifying drivers’ subjective perceptions and the objective visual landscape characteristics that influence them remains challenging due to the subjectivity of landscape perception, and the limitation of data size and the complexity of environmental information capture. This paper aims to overcome these limitations, based on deep learning techniques to achieve high-precision automatic identification and rapid processing of spatial visual elements in the highway landscape. Based on supervised fully connected neural networks, the ELO scoring algorithm, and ArcGIS, it realizes the quantitative evaluation of subjective perception emotions and visual mapping representation. Furthermore, the paper integrates optimization algorithms with deep learning, employing a Whale Optimization Algorithm (WOA) enhanced XGBoost regression model to analyze the interplay between spatial features and subjective experiences. The model’s effectiveness is confirmed through comparative experiments and SHAP analysis, offering insights into the coupling of subjective and objective data. The research results provide theoretical methods and quantitative analysis tools for the digitalization and refinement of highway landscape design, and also help researchers and urban planners to understand the interaction between subjective perception and objective semantics, providing new ways and insights for highway landscape planning and design.