语义分割技术下的城市滨水绿地美景度评价研究——以福州西湖公园、左海公园为例  

Landscape Evaluation on Urban Waterfront under Semantic Segmentation Technology—Taking Xihu Park and Zuohai Park in Fuzhou as Examples

作　　者：黎俊仪 林盈芳 董建文[1] 傅伟聪 LI Junyi;LIN Yingfang;DONG Jianwen;FU Weicong

机构地区：[1]福建农林大学风景园林与艺术学院,福州350002

出　　处：《中国园林》2022年第10期92-97,共6页Chinese Landscape Architecture

基　　金：福建农林大学高原学科建设项目“闽地人居环境可持续发展模式”(编号YSYL-kjtd-2)资助。

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。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.

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 

分 类 号：TU986[建筑科学—城市规划与设计]