文献类型：期刊文献

中文题名：基于街景图像的城市街道绿视率计量方法比较分析

英文题名：Comparison of Computational Methods for Urban Street Green View Indexes Based on Street View Images

作者：刘晓天[1,2] 孙冰[1] 廖超[1] 金佳莉[3] 施招婉[1] 范黎明[1] 唐艺家[1,2] 何继红[1,2] 何卫忠[4] 杨龙[5] 孙倩[6] 裴男才[1]

第一作者：刘晓天

机构：[1]中国林业科学研究院热带林业研究所,广东广州510520;[2]南京林业大学风景园林学院,江苏南京210037;[3]中国林业科学研究院林业研究所,北京100091;[4]广东省九连山林场,广东河源517100;[5]广州地理研究所,广东广州510070;[6]澳大利亚墨尔本皇家理工大学科学院地理空间科学系,维多利亚州墨尔本市VIC 3001

年份：2020

卷号：42

期号：5

起止页码：1022-1031

中文期刊名：江西农业大学学报

外文期刊名：Acta Agriculturae Universitatis Jiangxiensis

收录：CSTPCD;;北大核心:【北大核心2017】；CSCD:【CSCD2019_2020】；

基金：国家自然科学基金项目(31570594);广东省科学院建设国内一流研究机构行动专项资金项目(2020GDASYL-20200401001);广州市林业和园林局项目(2020-20);广东省林业发展及保护专项资金(2017-2018)。

语种：中文

中文关键词：绿视率;机器学习;卷积神经网络;街景图像;语义分割;城市林业;人居环境

外文关键词：green view index;machine learning;convolutional neural network;street scene image;semantic segmentation;urban forestry;human habitat

分类号：S73

摘要：【目的】针对街景图像在绿视率计量研究中的使用做出了说明,综合研究现状,对不同的绿视率数据获取方法和计算方法进行了介绍。【方法】对比了传统方法和利用算法模型(PSPNet或SegNet)的自动化方法,阐述了传统方法存在效率低、损耗大、精确度低等不足,而自动化方法有效的解决了这些问题。【结果】基于卷积神经网络模型的图像语义分割,使街景图片的处理与分析变得更为简便,但自动化方法仍有不足之处需要完善。【结论】文章指出利用机器学习来处理数据问题是未来研究发展的新趋势,预测融合机器学习和遥感技术的街景图像自动化计量方法将在城市规划建设、绿道效益评估等方面具有良好应用前景。
[Objective]By reviewing the application of street view imagery on the measurement of green view index(GVI),different methods for data acquisition and calculation were investigated.[Method]This paper compares the traditional method with the automation approach using the algorithm model(e.g.,PSPNet or SegNet),and explained the shortcomings of the traditional method,such as low efficiency,high loss and low accuracy,which can be effectively solved by the automation method.[Result]Through the semantic segmentation of convolution neural network model,the processing and analysis of street scenery pictures become simpler and efficient,though the new method still needs to be improved.[Conclusion]It is pointed out that using machine learning to process data is a new trend of future research and development;however,sufficient sample size and new algorithm are critical for robust GVI calculation.Finally,the paper proposes the combination of street view image with machine learning and remote sensing,and prospects practical application of the mentioned technologies in many fields such as urban planning and construction,and greenway benefit evaluation.