基于高分辨率遥感影像的城市建筑物目标识别

发布时间：2018-07-05 05:15

本文选题：特征提取 + 建筑物样本数据库　；参考：《北京建筑大学》2015年硕士论文

【摘要】：基于遥感影像处理和特征分析的方法,已经成为获取地面地物信息的主要方式之一,并且已经得到了广泛的应用和推广。二十世纪至今,遥感平台的迅速发展,遥感器多样性和精度的快速提高,使得分辨率在空间、光谱、辐射和时间上不断完善,在遥感影像上地物的信息也更加细腻丰富。因此,地面目标物的识别和特征提取的精度明显得到了改善。在城市高分辨率影像中,占据了八成左右的地面地物是建筑物和道路。故建筑物的识取占据了很大比率。城市建筑物对城市的运行、管理和规划有着重要的支撑作用。将来构建数字城市的关键基础技术之一就有城市建筑物的识别。因此研究建筑物的识别对城市的发展具有一定的意义。论文首先对SQL Server 2008数据库进行介绍,包括关系模型的特点、关系的规范化和数据库设计原则。接着对遥感影像实验区进行选取和描述,从光谱特征、几何特征和纹理特征对目标进行特点分析,提取出相关特征参数,并进行表结构设计,建立表关系,对数据进行入库存储,建立建筑物样本数据库。然后在基于数据库地物光谱特征的基础上,采用最小距离法、贝叶斯法、神经网络法和多分类法融合的基于不同权重综合分类法等监督分类方法对建筑物进行分类识别,并采用混淆矩阵对各种方法的分类结果做了精度评价和比较。最后选取出分类精度最高的影像,先用小图斑剔除法对影像进行后处理,接着以二值图像的方式提取出影像中的建筑物,并将小连通区域剔除,最后提取出建筑物边缘位置信息并标注在原始影像上。实验证明,建筑物样本数据库可以对城市建筑物的特征参数进行存储管理,对大城市建筑物多样性有较好的适应性,且该数据库可以为以其为基础的遥感影像建筑物的分类识别、空间分析提供可靠的基础数据。多分类法融合的基于不同权重综合分类法,根据各个分类法分类精度比重分配权重大小,有较强的适应能力,与人类对知识的认知习惯较符合。建筑物样本数据库的建立和多分类法融合的基于不同权重的综合分类器对于城市建筑物的识别有很好的效果,因此有一定的研究意义。
[Abstract]:The method based on remote sensing image processing and feature analysis has become one of the main ways to obtain ground ground information, and has been widely used and popularized. Since twentieth Century, the rapid development of remote sensing platform, the rapid improvement of the diversity and precision of remote sensing devices make the resolution in space, spectrum, radiation and time continuously. The information of ground objects on remote sensing images is also more delicate. Therefore, the accuracy of the recognition and feature extraction of ground objects has been improved obviously. In the urban high resolution images, about 80% of the ground objects occupy the buildings and roads. Operation, management and planning have an important supporting role. One of the key basic technologies for the future construction of digital cities is the identification of urban buildings. Therefore, it is of certain significance to study the identification of buildings for the development of the city. First, the paper introduces the SQL Server 2008 database, including the characteristics of the relationship model and the specification of the relationship. Then we select and describe the remote sensing image experimentation area, analyze the characteristics of the spectral features, geometric features and texture features, extract the related characteristic parameters, design the table structure, establish the table relationship, store the data and establish the database of the building sample. Then the data are based on the number of data. On the basis of the spectral characteristics of the storehouse, using the minimum distance method, the Bayesian method, the neural network method and the multi classification method to classify the buildings on the basis of the different weight comprehensive classification methods, and use the confusion matrix to evaluate and compare the classification results of various methods. Finally, the classification precision is selected. With the highest degree of image, the image is processed by the small plot elimination method first, then the building in the image is extracted with the two value image, and the small connected area is removed. Finally, the information of the building edge position is extracted and labeled on the original image. The experiment shows that the database of the building samples can be characteristic of the urban building. The parameters are stored and managed, and it has good adaptability to the diversity of the buildings in large cities. And the database can be used for classification and identification of remote sensing images based on it, and the spatial analysis provides reliable basic data. The multi classification method is based on the comprehensive classification of different weights, and the weight distribution right is classified according to the various classification methods. There is a strong ability to adapt to the cognitive habits of human knowledge. The establishment of the database of building samples and the integration of multiple classifications based on the multiple classifiers have a good effect on the identification of urban buildings, so there is a certain meaning of research.
【学位授予单位】：北京建筑大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：P237

【参考文献】