基于无人机影像的面向对象建筑物结构分级提取方法研究

(云南省地震局,云南 昆明 650224)

建筑物结构分类; 无人机高分辨率影像; 面向对象分类方法; 数字表面模型; 形状结构; 光谱特性

Research on Object-oriented Building Structure Classification Extraction Method Based on UAV Image
DU Haoguo,ZHANG Fanghao,LU Yongkun,CAO Yanbao,DENG Shurong,HE Shifang,ZHANG Yuanshuo,XU Junzu

(Yunnan Earthquake Agency,Kunming 650224,Yunnan,China)

building structure classification; high-resolution UAV image; object-oriented classification; DSM; shape structure; spectral characteristics

备注

为提高遥感影像建筑物结构识别精度,综合利用光谱、形状、空间、纹理和数字表面模型(DSM)建立了建筑物结构分级提取方法。基于研究区无人机高分辨率影像,采用面向对象的影像分析策略,首先进行多尺度分割,以最佳分割与合并指数提取影像中建筑物目标; 然后分别采用规则、训练样本与DSM方法对建筑物结构进行分类; 最后将3种分类方法进行融合,对比分析了单一方法和融合分类方法的建筑物结构分类精度。结果表明:基于规则+样本+DSM的半监督建筑物结构分类方法错分率、漏检率与Kappa系数最优。
In order to accurately recognize the building structure in remote sensing images,we proposed a method of extracting building structure at levels based on spectrum,shape,space,texture and digital surface model(DSM)[KG-*3].Then,by the help of high-resolution images taken by UAVs in the research area,we make object-oriented analysis.Firstly,we do multi-scale segmentation of the images in order to extract the objects according to the optimal segmentation and the merger index.Then,we classify building structures according to rules,training samples and DSM respectively.Finally,we combined the three classification methods,and get new results of the buildings structures.Then we compared the result from each method with the one from the combined method.We found that the semi-supervised classification based on rule+sample+DSM has the lowest error rate and omission ratio,and optimal Kappa coefficient.