遥感影像纹理分析及在地震灾害信息提取中的应用研究

发布时间：2018-11-04 17:49

【摘要】：纹理特征是遥感影像重要特性,纹理分析在遥感影像识别中的作用越来越大。大量研究表明,纹理分析是提高影像分类精度的重要手段,而影像分类的关键在于能否正确的提取影像纹理特征。如何快速、正确的提取影像纹理特征以及选择合适的特征维数一直热门的研究方向。本文基于典型地物在影像中的纹理特征,对影像纹理特征提取做了研究,根据提取出来的纹理特征值进行图像分类,并应用到地震灾害信息提取中,为震后救援提供信息参考。具体内容如下:首先在阅读大量前人研究成果基础上分析了典型地物如建筑物、山体、水体等在影像上的纹理特征,这些特征能够很好区分不同地物。总结了目前比较常用的纹理提取方法:统计分析方法、几何特征方法、信号处理方法和模型方法。分析了这些纹理分析方法的研究现状和各自特点,确定了统计分析方法中的灰度共生矩阵、行程长矩阵和灰度差分矩阵方法作为本文特征提取方法。编程实现特征提取算法,计算出图像的特征值。为了达到运算速度快、分类精度好的效果,对这几类方法的特征维数进行筛选,采用权重分析算法以及对比不同维数特征下的图像分类精度,确定本文图像分类实验采用15维特征量。选取玉树地震前后遥感影像作为实验样本,选取完好建筑、严重倒塌建筑和轻微倒塌建筑这三种能够代表地震灾害信息的样本进行分类实验,分类方法选用决策树算法和支持向量机分类器。分别进行单一地物分类和混合地物分类,单一地物分类样本选择建筑物,混合地物分类样本选择建筑物、山体、水体、道路等,实验结果表明本文采用的方法对建筑物分类正确率达到87%以上,能够从大量的样本中将建筑物识别出来。基于这些实验基础,进行灾害信息提取,直接在震后影像区域上循环对地物进行分类,识别出代表地震灾害信息建筑物样本。实验表明本文提取的方法能够较好的将地震信息提取出来,选择的特征量维数具有分类精度高,运行速度快的优点,比较具有代表性。
[Abstract]:Texture feature is an important feature of remote sensing image. Texture analysis plays a more and more important role in remote sensing image recognition. A large number of studies show that texture analysis is an important means to improve the accuracy of image classification, and the key of image classification is to extract the image texture features correctly. How to quickly and correctly extract image texture features and choose the appropriate feature dimension has been a hot research direction. Based on the texture features of typical ground objects in the image, this paper studies the extraction of image texture features, classifies the images according to the extracted texture features, and applies them to the extraction of earthquake disaster information, which provides information reference for post-earthquake rescue. The main contents are as follows: firstly, on the basis of reading a large number of previous research results, this paper analyzes the texture features of typical ground objects such as buildings, mountains, water bodies and so on, which can distinguish different features very well. The methods of texture extraction are summarized, such as statistical analysis method, geometric feature method, signal processing method and model method. The present situation and characteristics of these texture analysis methods are analyzed, and the gray level co-occurrence matrix, the travel length matrix and the gray difference matrix method are determined as the feature extraction methods in this paper. The feature extraction algorithm is realized by programming, and the feature value of the image is calculated. In order to achieve the effect of fast operation and good classification accuracy, the feature dimension of these methods is screened, and the image classification accuracy under different dimension features is compared by weight analysis algorithm. It is determined that the 15-dimensional feature quantity is used in the image classification experiment in this paper. The remote sensing images before and after the Yushu earthquake were selected as experimental samples, and three kinds of samples representing the information of earthquake disaster were selected, such as intact building, severely collapsed building and slightly collapsed building. Decision tree algorithm and support vector machine classifier are used for classification. The single feature classification and the mixed feature classification, the single feature classification sample select the building, the mixed feature classification sample selects the building, the mountain body, the water body, the road and so on. The experimental results show that the accuracy of the proposed method is more than 87% and can be recognized from a large number of samples. On the basis of these experiments, disaster information is extracted and ground objects are classified directly in the post-earthquake image area, and the building samples representing earthquake disaster information are identified. The experimental results show that the method can extract the seismic information well, and the dimension of the selected feature has the advantages of high classification accuracy and fast running speed, so it is more representative.
【学位授予单位】：中国地质大学(北京)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P315.7

【相似文献】