面向对象的高分遥感影像建筑物提取方法研究

发布时间：2018-07-24 11:24

【摘要】：建筑物是主要地理要素,是地表的重要组成和现代城市的发展标志,与人类生产生活关系密不可分,同时,建筑物也是遥感影像信息提取的重要研究目标之一。由于高分辨率遥感影像具有能够清楚反映海量地物细节信息,解析能力较强,覆盖范围更广,更新周期较快等明显特点,使用高分辨率遥感影像作为遥感信息提取的数据源能够取得较好效果,获取目标地物。因此,本文以面向对象的高分辨率遥感影像信息提取作为研究目标,以建筑物作为信息提取的研究对象,进行了面向对象的高分辨率遥感影像建筑物提取方法研究。主要研究成果及结论如下。(1)以SRG区域生长影像分割算法为基础,以提高分割效率、改善分割结果为目的,对区域生长中的生长准则和生长方式进行了研究。首先提出了利用高分辨率遥感影像的光谱信息计算区域生长中的生长准则,其次对SRG区域生长算法中的生长方式进行了改进,通过实验实现了高分辨率遥感影像分割,获取了影像对象。(2)影像对象是面向对象信息提取的基础。本文根据建筑物在高分辨率遥感影像上的具体表现,总结了建筑物对象的主要特征,包括建筑物在光谱、纹理、几何等方面的特征,建立了建筑物特征分类体系,并提出了一种提取高分辨率影像上影像对象的纹理-几何特征的方法。在特征提取方法中,首先改进了灰度共生矩阵计算方式,提取了能够更好地描述建筑物纹理信息的纹理特征,计算了相应的纹理统计量,同时提取了能够描述建筑物外形并有效地将建筑物与其他地物进行区分的几何特征,最后结合两种特征选择了8个具有代表性的特征值参数对影像对象尤其是建筑物进行综合描述。(3)提出了一种结合影像对象的纹理-几何特征与K均值分类思想的面向对象建筑物提取方法。该方法能够根据影像对象的特征及不同对象特征之间的差异对符合建筑物特性的特征进行分析判别,有效地将对象进行聚类,从而得到所需的建筑物信息。在实验部分选取了两组优秀的国内外高分辨率影像作为实验数据,包括国内的高分二号影像和国外的Pleiades高分辨率卫星影像,选择了基于规则的分类方法作为对比算法进行面向对象建筑物提取实验。首先利用本文分割方法提取了影像对象,其次计算了影像对象的纹理-几何特征,并将对象特征应用于K均值影像分类,最终成功地提取了实验影像上的建筑物信息,并对实验结果以及基于规则分类方法的实验结果进行了比较、分析和评价,证明了本文所提出的面向对象建筑物提取方法的有效性和可行性。
[Abstract]:Buildings are the main geographical elements, the important composition of the surface and the development symbol of modern cities, and are closely related to human production and life. At the same time, buildings are also one of the important research objectives of remote sensing image information extraction. Because the high-resolution remote sensing image can clearly reflect the massive detail information of the ground object, it has the advantages of strong analytical ability, wider coverage, faster updating period, and so on. Using high resolution remote sensing image as the data source of remote sensing information extraction can achieve good results and obtain target features. Therefore, the object oriented high resolution remote sensing image information extraction is taken as the research goal, and the building is taken as the research object of information extraction, and the object oriented high resolution remote sensing image building extraction method is studied in this paper. The main research results and conclusions are as follows: (1) based on the SRG region growing image segmentation algorithm, the growth criteria and growth patterns in the region growth are studied in order to improve the segmentation efficiency and the segmentation results. Firstly, the growth criterion of high resolution remote sensing image is proposed, then the growth mode of SRG region growth algorithm is improved, and the segmentation of high resolution remote sensing image is realized through experiments. The image object is obtained. (2) Image object is the basis of object oriented information extraction. According to the concrete performance of buildings in high-resolution remote sensing images, this paper summarizes the main features of building objects, including the spectral, texture and geometric features of buildings, and establishes a building feature classification system. A method for extracting texture-geometry features from high-resolution images is proposed. In the feature extraction method, we first improve the calculation method of gray level co-occurrence matrix, extract the texture features which can better describe the building texture information, and calculate the corresponding texture statistics. At the same time, the geometric features which can describe the shape of the building and effectively distinguish the building from other features are extracted. Finally, eight representative eigenvalue parameters are selected to describe the image object, especially the buildings. (3) an idea of combining texture-geometric feature and K-means classification of image object is proposed. Object-oriented building extraction method. Based on the features of image objects and the differences between the features of different objects, the method can analyze and distinguish the features that accord with the characteristics of buildings, and effectively cluster the objects to obtain the required building information. In the experiment part, two groups of excellent high-resolution images at home and abroad are selected as experimental data, including high-score 2 images in China and Pleiades high-resolution satellite images in foreign countries. The rule-based classification method is selected as the contrast algorithm for object-oriented building extraction experiment. Firstly, the image object is extracted by using the segmentation method in this paper. Secondly, the texture-geometry feature of the image object is calculated, and the object feature is applied to the classification of the K-means image. Finally, the building information on the experimental image is extracted successfully. The experimental results and the experimental results based on the rule classification method are compared, analyzed and evaluated, which proves the effectiveness and feasibility of the object-oriented building extraction method proposed in this paper.
【学位授予单位】：成都理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P237

【相似文献】