L型建筑物排列模式的深度学习识别方法  

Deep Learning Recognition Method of L-shaped Building Arrangement Pattern

作　　者：袁远 孟妮娜[1] Yuan Yuan;Meng Nina(School of Geodesy and Surveying and Mapping,Chang'an University,Xi'an 710054,China)

机构地区：[1]长安大学地质工程与测绘学院,陕西西安710054

出　　处：《甘肃科学学报》2021年第3期7-11,共5页Journal of Gansu Sciences

基　　金：国家自然科学基金项目(41501498);长安大学中央高校基本科研业务费专项资金项目(300102260403);地理信息工程国家重点实验室自主研究课题资助项目(SKLGIE2019-ZZ-2)。

摘　　要：建筑物空间分布模式表现为建筑物群组在空间中所体现出的明确的组织结构,并能够在人类视知觉认知过程中被识别和命名。建筑物空间分布模式的探测和识别对于制图综合具有重要意义。针对传统的建筑物模式识别方法中各参数及阈值确定困难的问题,文中以L型排列为例,设计了一种基于图卷积神经网络的识别方法。该方法构建了矢量建筑物数据的图结构,以各建筑物自身的特征及其排列的结构特征作为输出,搭建了图卷积神经网络对样本进行监督学习并得到识别模型。最后利用OpenStreetMap公开数据集进行验证。实验结果表明,该方法能够有效地识别出L型建筑物排列。The spatial distribution pattern of buildings is the clear organization structure of building groups in space,which can be recognized and named in the process of human visual perception.The detection and recognition of building spatial distribution pattern is of great significance for cartographic generalization.In view of the difficulty in determining parameters and thresholds in traditional building pattern recognition methods,this paper designs a recognition method based on graph convolution neural network by taking L-shaped arrangement as an example.In this method,the graph structure of vector building data is constructed,and the structural features of each building and its arrangement are taken as the output.The graph convolution neural network is built to supervise the learning of samples and get the prediction model.Finally,OpenStreetMap is used to open the dataset for verification.The experimental results show that this method can effectively identify the L-shaped building arrangement.

关 键 词：地图综合 模式识别 建筑物群 分布模式 

分 类 号：P208[天文地球—地图制图学与地理信息工程]