Using explainable machine learning and image generation algorithm to objectively calculate and intelligently optimize the aesthetic quality of rural road environments

Artikel i vetenskaplig tidskrift, 2026

With the development of self-drive tourists in rural areas, the aesthetic quality of rural road environments has become increasingly important. However, current methods of aesthetic calculation for road environments are highly subjective, and the optimization methods rely heavily on expert experience. Thus, this study seeks to objectively calculate rural road environments’ aesthetic quality using explainable machine learning and conduct intelligent optimization using an image generation algorithm. Utilizing naturalistic driving data, road environment elements (i.e., landscape, color, and texture information) are extracted, and the aesthetic features (i.e., Variety, Unity, and Symmetry) of these elements are objectively quantified. Combining XGBoost with SHAP, an objective and explainable aesthetic calculation model of rural road environments is proposed to calculate aesthetic scores and interpret contributions of road environment elements to aesthetic features. Based on these scores and interpretable insights, optimization objects are effectively identified. Subsequently, corresponding text prompts are gained and input into Stable diffusion model for the automatic generation of optimized rural road environment images. The results indicate superior performance of the proposed aesthetic calculation model compared to other commonly-used machine learning methods, demonstrating an average accuracy enhancement by at least 20%. Case studies prove that intelligent optimization based on Stable diffusion model can improve the overall aesthetic score and aesthetic feature score by an average of 64.5% and 121.5%, respectively. Compared with another popular image generation algorithm CycleGAN, Stable diffusion model provides better optimization results. Findings in this study can help to objectively and intelligently improve the aesthetic quality of rural road environments.