基于网络照片数据与Auto ML模型的湖泊公园景观意象特征及感知偏好研究

景观意象作为游客主体和景观客体之间的映射，是提升景观吸引力重要的理论依据。以武汉市的7个湖泊公园为研究对象，通过Google Cloud Vision提供的在线机器学习算法，自定义景观标签，构建Auto ML模型对16883张网络景观照片进行图像挖掘与内容识别，归纳景观意象的具体维度，从景观类型、景观要素、空间尺度和意象元素4个层面分析公众对景观意象的感知特征，探讨不同样点的感知共性与感知特性以及影响偏好差异的原因。结果表明：湖泊公园中自然景观的感知高于人文景观，其中水体景观、林木景观、历史文化是湖泊公园的核心感知要素；大尺度空间广阔深远的自然景观和小尺度精细化设计的人文景观更受游客偏好；常绿树、湖泊、背景建筑群、草地和倒影是湖泊公园中感知最强的意象元素；以及体现公园特色的特异性景观更容易得到游客的高度感知。本研究丰富了景观意象的研究内容与分析方法，为以大数据图片实现景观偏好的相关研究提供了思路与参考。

A Study on Landscape Image and Public Perception and Preferences of Lake Parks Based on Internet Photo Data and Auto ML Model

The study of landscape image serves as an important theoretical basis for improving landscape attractiveness which correlates the subject of tourists and the object of landscape. This study used the online machine-learning algorithm provided by Google Cloud Vision to categorize 16 883 landscape photos of seven lake parks in Wuhan with the Auto ML model trained by self-developed landscape tags, so as to conduct image mining and content recognition. Then people''s perceptions of different types and elements of landscape, space dimensions and imagery elements were analyzed to explore how common or different people''s preferences were towards these landscape images as well as the factors affecting people''s different perceptions. The results show that: 1) the natural landscape of lake parks is better appreciated than man-made landscape, among which waters, trees as well as historical and cultural elements are the core factors affecting people''s perceptions; 2) natural landscape covering a large area with a broad view and man-made landscape with delicate designs covering a relatively small area are preferred by tourists; 3) evergreen trees, lakes, background buildings, grass and water reflection are the most appreciated imagery elements; and 4) landscape with unique natural or man-made units are more likely to be highly appreciated by tourists. Therefore, this study provides new observations on the subjects and analytical methods of landscape image, and presents some references for big data analysis of photos and landscape preference.