Abstract: The transformation of traditional Chinese building facades plays a crucial role in enhancing urban aesthetics and cultural significance. Scientifically evaluating the visual effects of commercial renovations on traditional Chinese building facades is of paramount importance for both theoretical understanding and practical implementation. Conventional research methods often rely on manual questionnaire surveys and manual data analysis, which are not only limited in terms of cost, time, and measurement scale but also susceptible to the subjective preferences of respondents and individual differences. In this study, an image dataset containing 560 images of commercial remodelling of traditional building facades was first constructed. Based on this dataset, a deep learning-based classification model, Swin-HV was developed, which can evaluate and predict the visual effect of façade renovation in terms of both historical and cultural atmosphere and visual preference. Secondly, a YOLOv8-based object detection method was used to identify nine object categories from the commercial renovation images of traditional building facades, and a multiple linear regression model was used to analyse the correlation between the architectural elements and the evaluation of visual effects. Additionally, Grad-CAM++ was utilized to visualize the decision-making process of the model. The results demonstrate that the Swin-HV model achieves high accuracy in predicting evaluations of historical cultural ambiance and visual preferences. Moreover, the study revealed a close relationship between the visual effects evaluation of commercial renovations on traditional building facades and architectural elements. The methodology proposed in this study provides insights for urban planning and architectural preservation, deepening our understanding of commercial renovations on traditional building facades.

traditional building renovation, deep learning, visual preference