基于LiDAR点云与高分影像的面向对象的损毁建筑物提取方法研究

发布时间：2018-03-06 19:19

本文选题：LiDAR　切入点：高分影像　出处：《西南交通大学》2013年硕士论文　论文类型：学位论文

【摘要】：针对地震灾区地形复杂,地物光谱特征、纹理特征与空间分布复杂,利用单一数据源无法快速、准确地提取损毁建筑物信息等情况,本文利用高分影像所具有的光谱特征、纹理特征和LiDAR点云所具有的高精度的空间三维信息、二次回波信息应用面向对象的影像分类技术,以实现损毁建筑物的快速、准确提取为目的,开展了以下研究工作： 1、归纳总结了现有LiDAR点云滤波分类方法、LiDAR点云与高分影像的配准方法。采取人机交互的方式,利用基于面元特征的配准方法实现了LiDAR点云和高分影像的配准。 2、综述了面向对象的高分影像分割分类方法,主要包括：高分影像的多尺度分割方法、最优分割尺度确定以及面向对象的模糊分类方法。 3、研究采用LiDAR点云以nDSM的方式参与多尺度分割,实验表明nDSM作为LiDAR点云的“高程”信息参与多尺度分割并不会造成影像的“过分割”现象,而是加速了影像对象的合并过程。 4、本文提出最优分割尺度范围值求交、取最小值得到最优分割尺度的方法,该方法优于原有的最优分割尺度确定方法。实验表明该方法是可行的,该方法具有更广的适用范围。 5、研究分析了实验研究区域内各地物目标的光谱特征、形状特征、纹理特征等,通过样本的选择、实验分析选定了各地物目标的分类特征,建立了模糊分类规则集。根据树木在高分影像上的光谱特征构造了新的植被指数(VInew),实验表明：利用该指数能够准确有效的提取出树木信息,提取树木信息的Kappa系数为0.981744。 6、根据损毁建筑物的特征,采用通过逐步剔除影像上的树木、道路、建筑物的方式提取损毁建筑物。最后对面向对象的损毁建筑物提取结果进行了精度分析,实验表明该方法切实可行,提取损毁建筑物Kappa系数为0.932576。综上,本文研究的基于LiDAR点云与高分影像的面向对象的损毁建筑物提取方法是切实可行的,损毁建筑物提取结果的用户精度为94.04%,生产者精度为95.30%,Kappa系数为0.932576。本文的研究成果对于实现损毁建筑物的快速、准确提取具有一定的参考价值,为灾后建筑物损毁评估、应急响应的信息获取提供了可靠的技术支持。
[Abstract]:In view of the complex terrain, spectral features, texture features and spatial distribution of the quake-hit areas, the information of damaged buildings can not be extracted quickly and accurately by using a single data source, and the spectral features of high-score images are used in this paper. Texture feature and LiDAR point cloud have high precision spatial 3D information. The secondary echo information applies object oriented image classification technology, in order to realize the fast and accurate extraction of damaged buildings, the following research work is carried out:. 1. The existing LiDAR point cloud filtering classification methods are summarized, and the registration method of LiDAR point cloud and high score image is realized by using the method of human-computer interaction. 2. The object-oriented high-score image segmentation and classification methods are summarized, including multi-scale segmentation method, optimal segmentation scale determination and object-oriented fuzzy classification method. 3. Using LiDAR point cloud to participate in multi-scale segmentation by means of nDSM, the experiment shows that nDSM, as the "elevation" information of LiDAR point cloud, does not result in "over-segmentation" of image, but accelerates the merging process of image objects. 4. In this paper, the method of finding the intersection of the range value of the optimal segmentation scale and obtaining the optimal partition scale by taking the minimum value is proposed. The method is superior to the original method for determining the optimal segmentation scale. The experiment shows that the method is feasible and has a wider range of application. 5. The spectral feature, shape feature, texture feature and so on of the objects in the experimental study area are analyzed. Through the selection of samples, the classification features of the objects are selected. A fuzzy classification rule set was established, and a new vegetation index was constructed according to the spectral characteristics of trees in high score images. The experimental results show that the Kappa coefficient of tree information can be extracted accurately and effectively by using this index, and the Kappa coefficient of extracting tree information is 0.981744. 6. According to the characteristics of damaged buildings, the damaged buildings are extracted by gradually removing the trees, roads and buildings on the image. Finally, the precision analysis of the result of the object oriented damaged building extraction is carried out. The experimental results show that this method is feasible and the Kappa coefficient of damaged buildings is 0.932576. In summary, the method of object oriented damage building extraction based on LiDAR point cloud and high score image is feasible. The user accuracy of the damaged building extraction results is 94.04 and the producer accuracy is 95.30 and the Kappa coefficient is 0.932576.The research results in this paper have certain reference value for the rapid and accurate extraction of damaged buildings, and are useful for the assessment of building damage after the disaster. The information acquisition of emergency response provides reliable technical support.
【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：P225.2

【参考文献】