基于U2⁃net深层网络架构的古镇街道街景图像绿化质量测度研究  

Research on measurement of greening quality of street view image of ancient town based on U2-net deep network architecture

作　　者：张春明[1] 谭人殊[1] 程海帆 ZHANG Chunming;TAN Renshu;CHENG Haifan(Design School,Yunnan Art University,Kunming 650500,China;School of Architecture and Urban Planning,Kunming University of Technology,Kunming 650504,China)

机构地区：[1]云南艺术学院设计学院,昆明650500 [2]昆明理工大学建筑与城市规划学院,昆明650504

出　　处：《中南民族大学学报（自然科学版）》2023年第3期357-364,共8页Journal of South-Central University for Nationalities：Natural Science Edition

基　　金：教育部人文社会科学研究青年基金资助项目(21YJC760070)。

摘　　要：街道绿视率即街景图片中绿色面积在街景图片中的占比值,借此数值来评价人本感知视角下的街道绿化品质.针对这一目标,基于开源街景数据和机器学习算法,提出了一套人本视角下街道语义划分和测度方法.以大理古城为例,快速且高效地实现了该范围内街道语义的计算与可视化.采用U2-net神经网络模型进行街景语义数据分析,以“树”语义为代表,在数据分析的基础之上提出了新的测算方式,实现对于整体街道的绿视率量化评测,并在此基础之上提出了相应的设计更新策略.The street green visibility is the ratio of green area and street view area in the street view picture,which is used to evaluate the street greening quality from the perspective of human perception.Aiming at this goal,based on open-source street view data and machine learning algorithm,a set of street semantic division and measurement methods is proposed from a humanistic perspective.Taking the ancient city of Dali as an example,itrealizes the calculation and visualization of street semantics within this range quickly and efficiently.U2-net neural network model is used to analyze the semantic data of the street scene.Represented by the“tree”semantics,a new measurement method is proposed on the basis of data analysis to achieve the quantitative evaluation of the overall street green viewing rate,and on this basis,the corresponding design update strategy is proposed.

关 键 词：深度学习 优化措施 街景 绿化 

分 类 号：TU984.2[建筑科学—城市规划与设计] TP181[自动化与计算机技术—控制理论与控制工程]