基于倾斜摄影综合SVM与模糊规则的城市不透水面信息提取

Urban Impervious Surface Extraction by Integrating SVM and Fuzzy Rules Based on Oblique Photography

准确获取城市不透水面信息对于城市建设规划、环境评估等具有重要意义。该文建立了一种基于SVM监督分类和模糊规则分类相结合的城市不透水面提取方法,首先基于研究区无人机航摄多视倾斜影像,采用SFM方法重建三维点云数据,生成该区域的DOM、DEM和nDSM;其次,将研究区的波段信息和DEM、nDSM数据等共同作为分类依据,基于面向对象的多尺度图像分割技术,融合光谱波段和DEM、nDSM等数据,获取分割对象及其特征;最后,综合利用SVM分类与基于模糊规则的分类方法对城市不透水面信息进行提取。实验结果表明,该方法整体精度为0.91,Kappa系数为0.89,优于单独采用基于SVM的分类方法或基于模糊规则的分类方法。

The coverage of impervious surface is an important factor to measure the development of cities and ecological environment.It is very important to obtain the impervious surface information on urban construction planning and environmental assessment.The point cloud data extracted from the UAV oblique photography image contains detailed 3D information of the surface.This paper established a technical process of urban impermeable surface extraction based on SVM supervised classification and fuzzy rule classification.Firstly,the paper reconstructed the 3D point cloud data by UAV multi-view images based on SFM method and generated the DOM,DEM and nDSM models in the study area.The three visible light bands of R,G,B of DOM image and the values of DEM and nDSM were taken as the basis of classification.By using object-oriented multi-scale segmentation technology,the impermeable information in the study area was extracted based on SVM supervised classification method,fuzzy rule classification method,and integrated SVM and fuzzy rule classification method,respectively.The experimental results show that the overall accuracy of the integrated SVM and fuzzy rule classification is 0.91 and the Kappa coefficient is 0.89,which is better than the SVM supervised classification method or the fuzzy rule classification method.