A New Approach to Landscape Visual Quality Assessment from a Fine-Tuning Perspective

Various methods for evaluating the visual quality of landscapes have been continuously studied. In the era of the rapid development of big data, methods to obtain evaluation data efficiently and accurately have received attention. However, few studies have been conducted to optimize the evaluation methods for landscape visual quality. Here, we aim to develop an evaluation model that is model fine-tuned using Scenic Beauty Evaluation (SBE) results. In elucidating the methodology, it is imperative to delve into the intricacies of refining the evaluation process. First, fine-tuning the model can be initiated with a scoring test on a small population, serving as an efficient starting point. Second, determining the optimal hyperparameter settings necessitates establishing intervals within a threshold range tailored to the characteristics of the dataset. Third, from the pool of fine-tuned models, selecting the one exhibiting optimal performance is crucial for accurately predicting the visual quality of the landscape within the study population. Lastly, through the interpolation process, discernible differences in landscape aesthetics within the core monitoring area can be visually distinguished, thereby reinforcing the reliability and practicality of the new method. In order to demonstrate the efficiency and practicality of the new method, we chose the core section of the famous Beijing–Hangzhou Grand Canal in Wujiang District, China, as a case study. The results show the following: (1) Fine-tuning the model can start with a scoring test on a small population. (2) The optimal hyperparameter setting intervals of the model need to be set in a threshold range according to different dataset characteristics. (3) The model with optimal performance is selected among the four fine-tuning models for predicting the visual quality of the landscape in the study population. (4) After the interpolation process, the differences in landscape aesthetics within the core monitoring area can be visually distinguished. We believe that the new method is efficient, accurate, and practically applicable for improving landscape visual quality evaluation.