| 摘要: |
| 为探讨城市公园滨水绿地景观与公众偏好之间的关
系，利用景观评价法结合图像语义分割、灰色统计分析等方
法，对福州市西湖公园、左海公园等36个典型城市滨水绿地
进行了研究，应用虚拟现实技术结合问卷访谈获取公众审美
评价，并利用图像语义分割技术实现要素量化分解，以此展
开城市滨水绿地美景度评价研究。结果表明：1)PSPNet模
型对城市滨水绿地的分割精度为88.8%，适用于该场景类型
的分析；2)美景度模型结果显示具高贡献率的5个主要因子依
次是商业设施(27.8%)、视觉复杂度(25.8%)、植物生长状
况(10.8%)、植物层次(10.4%)、铺装形式(8.4%)，商业
设施和视觉复杂度对美景度产生负向影响，植物生长状况、
植物层次、铺装形式则正向促进公众审美感知；3)高美景度
评价(SBE)得分场景特征为：绿视率为30%~50%，蓝视率为
2%~10%，铺装比例为15%~30%；低美景度评价(SBE)得分
场景最突出特征为商业氛围浓厚(4.19%)。所得结果可为后
续滨水绿地的景观规划设计及管理提供客观数据参考。 |
| 关键词: 风景园林 城市滨水 虚拟现实 图像语义分
割 美景度评价法 |
| DOI： |
|
| 基金项目: |
|
| Landscape Evaluation on Urban Waterfront under Semantic Segmentation Technology—Taking Xihu Park and Zuohai Park in Fuzhou as Examples |
| LI Junyi,LIN Yingfang,DONG Jianwen,FU Weicong |
| Abstract: |
| To explore the relationship between the waterfront
green space landscape of urban parks and public preferences,
the landscape evaluation method combined with Semantic
Segmentation, grey statistical analysis and other methods were
used to analyze 36 typical waterfront green spaces in Xihu Park
and Zuohai Park in Fuzhou. Virtual Reality technology combined
with questionnaire interviews are used to obtain public aesthetic
evaluation, and semantic segmentation technology is used to
achieve quantitative decomposition of landscape elements, so as to
study the evaluation of the beauty of urban waterfront green space.
The result shows: 1) PSPNet model has a recognition accuracy of
88.8% for urban waterfront green space, which is suitable for the
analysis of this type of scene; 2) the results of the beauty model
show that the five main factors with high contribution rates are
commercial facilities (27.8%), visual complexity (25.8%), plant
growth status (10.8%), plant level (10.4%), and paving form (8.4%).
Commercial facilities and visual complexity have a negative impact
on beauty, while plant growth status, plant level, and paving form
positively promote public aesthetic perception; 3) the characteristics
of scenes with high SBE (scenic beauty estimation) values are:
green view rate is 30%-50%, blue view rate is 2%-10%, and the
pavement ratio is 15%-30%; the most prominent feature of low
SBE value scenes is the strong commercial atmosphere. The results
can provide objective data reference for the subsequent landscape
planning, design and management of waterfront green space |
| Key words: : landscape architecture urban waterfront virtual reality semantic segmentation scenic beauty estimation |
