Abstract
Mountainous scenic spots are important tourism resources, and the study of tourists' landscape perception and emotional preference when visiting them is beneficial to the management of scenic spots in order to improve the service quality and promote the protection, development, and utilization of scenic resources. In this paper, we use the location photo data of tourists at Huangshan Mountain to extract the visual semantic information of location photos, calculate the photo sentiment value, and mine the landscape perception and sentiment preference features of tourists using DeepSentiBank image recognition model and photo visual semantic quantification method. The results show the following: (1) Huangshan tourists mainly focus on nine types of photos, with the most attention paid to the category of mountain rock landscapes and the least attention paid to the category of animal landscapes. (2) In terms of spatial distribution, the landscape types of tourist photos show the spatial characteristics of "concentrated into a belt", "significant nucleus", and "fragmented distribution". The spatial variation of the emotional value of tourists' photos is significant, and the high values are mainly distributed at the entrances and exits, interchanges, and famous attractions. (3) On a temporal scale, the type of perception of the Huangshan location photograph landscape shows a significant imbalance. The emotional values of tourists' photos vary significantly, with a "slowly sloping straight line" type of emotional change on the seasonal scale, a "W" type of emotional change on the monthly scale, an "N" type of emotional change on the weekly scale, and an "M" type of emotional change on the hourly scale. This study attempts to explore the landscape perceptions and emotional preferences of tourists in mountainous scenic areas with new data and methods, aiming to promote the sustainable and high-quality development of mountainous scenic areas.