您好, 访客   登录/注册

基于GIS和历史卫星影像的城市建筑大数据识别

来源:用户上传      作者:邓章 陈毅兴

  摘 要:在城市建筑群能耗模拟中,建筑类型和建筑年代是典型建筑参考的主要依据,目 前较难直接获取相关数据.为识别建筑类型,以长沙市区21538个建筑轮廓(不含城市地图信 息点和区域边界轮廓信息)为例,基于建筑轮廓的轮廓面积、近似矩形短边宽度、近似矩形系数等几何特征,运用随机森林方法成功识别出低层住宅、公寓式住宅和其他类型,整体准确率为81.7%.为识别建筑年代,以长沙市中心区域 7900个建筑轮廓为例,基于历史卫星影像数据,运用卷积神经网络方法自动提取不同年代的建筑轮廓,平均精确度为80%.然后分别相交分析推断出 5077 栋建筑建造于2005年之前,1606 栋建筑建造于2005―2014年,1217 栋建筑 建造于2015―2017年.该方法同样适用于其他城市,为后续的建筑群能耗模拟提供了数据 支持.
  关键词:城市建筑群能耗模拟;建筑类型;建造年代;随机森林;卷积神经网络中图分类号:TU111 文献标志码:A
  Identification of City-scale Building Information Based on GIS Datasets and Historical Satellite Imagery
  DENG Zhang1,CHEN Yixing1,2?
  (1.College of Civil Engineering,Hunan University,Changsha 410082,China;
  2.Key Laboratory of Building Safety and Energy Efficiency of the Ministry of Education(Hunan University),Changsha 410082,China)
  Abstract:Building type and built year are critical parameters to infer archetype buildings for urban building en-ergy modeling(UBEM).Currently,it is difficult to directly obtain these data for most cities.For the building type identification,taking 21538 building footprints(without a point of interest and community boundary information)in Changsha City as an example,this paper used the random forest algorithm to successfully identify low-rise resi-dences,apartment residences,and other types based on the geometric characteristics,with an overall accuracy of81.7%.For the determination of built year,7900building footprints in the downtown area of Changsha were selected as a case study,and this paper applied the convolutional neural network algorithm to automatically extract building footprints from different historical satellite imageries,with an average precision of80%.Then,the intersection analy-sis showed that 5077 buildings were built before 2005,1606 buildings were built from 2005 to 2014,and1217buildings were built from 2015 to 2017.The proposed method can be easily applied to other cities,and provide data support for UBEM in the future.
  Key words:urban building energy modeling;building type;built year;random forest;convolutional neural net-work
  S着城镇化进程加快,建筑能耗总量不断上升,建筑成为第三“能耗大户”[1],因此建筑节能对城市的可持续发展尤为关键.建筑能耗模拟可用于评价 节能技术措施[2].城市尺度的建筑群能耗模拟是国 际城市能源研究领域的一个新兴方向,可以更好地 评估新区能源规划和旧区节能改造等技术方案,从而推动节能减排目标的实施.由于缺乏每栋建筑的详细数据,在城市建筑群能耗模拟中,围护结构和空调系统等参数一般根据典型建筑进行假定,而建筑类型及建造年代是典型建筑参考的主要依据[3].
  目前获取数据最直接的方法是利用政府机构公开的数据平台,绝大部分的研究中都采用这种方式.欧美一些大城市的数据平台存储了大量城市建筑信 息,如建筑轮廓、楼层数、建筑类型和建造年代等数据[4-5],可用于建筑群能耗模拟.公开数据平台节省了大量收集数据的时间,但受限于特定的城市.另一种直接的方法是实地调研[6].当调研的范围扩大至城市级别,是极其耗时耗力的.

nlc202208221000



转载注明来源:https://www.xzbu.com/4/view-15438022.htm

相关文章