摘要
因传统方法单一使用遥感影像或兴趣点(point of interest,POI)数据识别城市功能区时,存在城市特征信息利用不完全、识别精度不高的问题,提出利用POI数据、遥感影像等多源数据,并将自然特征和人文特征相结合,采用基于掩膜区域卷积神经网络和样方密度法(mask region based convolutional neural network and quadrat density method,Mask R-CNN-QDM)模型识别城市功能区的方法。首先基于遥感影像采用Mask R-CNN模型识别建筑物,然后将识别结果与POI数据进行补充校验,得到结合自然特征和人文特征的分类结果,再引入面积要素对分类结果进行赋分,以计算样方密度,并采用随机抽样方式对所提方法功能区的识别精度进行评价。研究结果表明,Mask R-CNN-QDM模型的识别精确度高达0.900,平均Kappa系数为0.802,说明该方法能较好地区分单一城市功能区和混合城市功能区。
Aiming at the problem of incomplete utilization of urban feature information and low recognition accuracy with traditional single use of remote sensing images and point of interest(POI)data to identify urban functional areas,this paper proposes a mask region based convolutional neural network and quadrat density method(Mask R-CNN-QDM)for identifying urban functional areas by combining natural and cultural features using multi-source data such as POI data and remote sensing images.Firstly,Mask R-CNN is used for identifying buildings from remote sensing images,and then the extraction results are supplemented and verified with POI data to obtain classification results that combines natural and cultural features.Secondly,area factor is introduced to divide classification results and calculate quadrat density.Finally,the accuracy of the proposed method is evaluated by using random sampling to identify the functional areas.The research results show that recognition accuracy of Mask R-CNN-QDM is as high as 0.900,and the average Kappa coefficient is 0.802,indicating that this method can effectively distinguish between single urban functional areas and mixed urban functional areas.
作者
牛成英
邢晓文
闫新宇
NIU Chengying;XING Xiaowen;YAN Xinyu(School of Statistics,Lanzhou University of Finance and Economics,Lanzhou 730020,China)
出处
《湖北民族大学学报（自然科学版）》
CAS
2023年第3期405-413,共9页
Journal of Hubei Minzu University:Natural Science Edition
基金
国家社会科学基金项目(21BTJ042)
中央引导地方科技发展项目(YDZX20216200001876)
甘肃省高等学校创新基金项目(2023B-091)。
