Research on rural landscape preference evaluation using deep learning

Rural roads play a vital role in connecting residents and supporting local production, influencing the overall aesthetic of villages. Despite their significance, existing rural research often overlooks the landscapes along these roads and lacks a standardized system for quantifying visual quality. Our study endeavors to fill this knowledge gap by harnessing computer vision technology to create a semantic segmentation model tailored for rural landscapes and to formulate a systematic, quantifiable index of landscape features. Employing a semantic image segmentation algorithm, we extracted physical landscape characteristics from panoramic images sourced from Changsha County. Through meticulous examination of the correlation between these measured landscape attributes and visual perception via multiple linear regression analysis, we achieved remarkable accuracy in our semantic segmentation model, reaching a precision of 83 percent. The results from the regression models highlighted the significant impact of the green vision index, farmland visibility index, building density index, and road width index on visual perception. Specifically, green vision index and farmland visibility index were positively correlated, while building density index and road width index showed negative correlations. In rural landscape planning, optimizing the balance between preserving and enhancing natural landscapes and incorporating practical, visually appealing artificial features is essential. This study presents a pioneering research framework that harnesses computer vision technology to conduct detailed and comprehensive rural landscape analyses, thereby furnishing a quantitative means of evaluating extensive rural landscapes.