基于Hyperion高光谱数据的城市地物识别与分类研究

发布时间：2018-03-05 09:36

本文选题：Hyperion高光谱遥感　切入点：城市地物　出处：《浙江大学》2013年硕士论文　论文类型：学位论文

【摘要】：城市化飞速发展的今天,对于城市环境信息的监测对于改善城市生态环境、规范城市规划管理等具有重要的意义。城市下垫面尤其是大量不同年代、材料、成分的人工地物,其光谱多样性远超过自然环境。高光谱数据丰富的光谱信息可以弥补传统遥感数据源光谱分辨率方面的不足,从而实现对城市地物更为精细的识别和分类。对此,本文从以下几个方面对高光谱城市地物识别进行了探讨：首先,本文阐述了采用高光谱数据进行城市研究的意义和目标；介绍了高光谱遥感硬件的发展概况,概括了大气校正技术、光谱特征提取、影像融合及地物识别和分类技术等影像分析技术的研究动态,以及高光谱遥感在地质调查、植被分析、水环境监测、农业信息和大气环境等领域的应用；提出里本次研究的主要内容和研究框架。其次,对本次研究的区块现状进行介绍,针对Hyperion高光谱数据,通过几何校正、辐射定标、波段选择以及FLAASH (Fast Line-of-sight Atmospheric Analysis of Spectral Hyper-cubes)大气校正消除Smile效应等一系列预处理,获得地物的真实反射率；再根据研究区内几种典型地物在全波段范围内的光谱特性以及不同波段的信息量和相关性,对波段进行重采样,保留信息含量多、相关性小和地物可分性强的波段作为最佳波段。在此基础上,通过总结归纳遥感影像数据融合的发展现状及各常用算法的优缺点,采用Gram-Schimdt (GS)正交化变换法,以高分辨率的SPOT全色影像为基准影像,对高光谱数据进行融合处理,融合后的影像在空间分辨率上有明显的提高,并且地物的光谱信息损失不大,保持了原有的光谱形态。再次,对城市常见地物类型的光谱特征进行分析,根据研究区实际情况,通过实地调查及遥感影像目视解译,确定九类城市地物作为研究对象。在此基础上针对现有地物端元提取方法的不足,采用纯净像元指数与光谱角匹配(SAM, Spectral Angle Mapper)相结合的方法提取了九类地物的端元光谱并建立参考光谱库,作为后续地物识别分类的基础。最后,针对现有常用的高光谱影像识别及分类方法,采用监督分类中的光谱角匹配方法(SAM)和线性光谱分解方法(LSU, Linear Spectral Unmixing)分别对最佳波段选择前后及数据融合前后的高光谱影像进行识别分类,并进行图像结果及地物面积统计分析。结果表明：星载高光谱数据可以较为准确的识别出常见的城市地物类型,采用的识别方法对结果尤为重要,并且高光谱影像的融合处理可以一定程度上提高分类结果的精细度和准确度；当采用SAM方法对融合后的影像进行识别时,其地物面积统计误差仅为11.61%,而采用LSU方法对未经融合的影像进行识别,其误差将达到65.63%,并且其图像结果浑浊不清,无法分辨各个地物类型的分布及聚集形式。
[Abstract]:With the rapid development of urbanization, the monitoring of urban environmental information is of great significance for improving the urban ecological environment and standardizing urban planning and management. The spectral diversity is far greater than that of the natural environment. The spectral information rich in hyperspectral data can make up for the deficiencies in spectral resolution of traditional remote sensing data sources, thus achieving more precise recognition and classification of urban features. This paper discusses the recognition of hyperspectral urban features from the following aspects:. Firstly, the significance and goal of using hyperspectral data in urban research are described, and the development of hyperspectral remote sensing hardware is introduced. The atmospheric correction technology and spectral feature extraction are summarized. The research trend of image analysis technology, such as image fusion and ground object recognition and classification, and the application of hyperspectral remote sensing in geological survey, vegetation analysis, water environment monitoring, agricultural information and atmospheric environment. Put forward the main contents and research framework of this study. Secondly, the current status of the blocks in this study is introduced. For Hyperion hyperspectral data, a series of preprocessing processes such as geometric correction, radiometric calibration, band selection and FLAASH Fast Line-of-sight Atmospheric Analysis of Spectral Hyper-cubes-based atmospheric correction to eliminate Smile effect are presented. According to the spectral characteristics of several typical ground objects in the study area and the amount and correlation of information in different bands, the true reflectivity of the ground objects is obtained. The band with small correlation and strong separability is the best band. On this basis, by summing up the development of remote sensing image data fusion and the advantages and disadvantages of the common algorithms, we adopt the Gram-Schimdt / GSH) orthogonal transform method, and take the high-resolution SPOT panchromatic image as the reference image. The fusion of hyperspectral data shows that the spatial resolution of the fused image is improved obviously and the spectral information of the ground object is not lost so that the original spectral form is maintained. Thirdly, the spectral characteristics of common urban features are analyzed, according to the actual situation of the study area, through field investigation and remote sensing image visual interpretation, Nine kinds of urban features are determined as the object of study. Based on this, the shortcomings of the existing methods for extracting endpoints of ground objects are pointed out. Using the method of pure pixel index and Spectral Angle Mapper, the end-element spectra of nine kinds of ground objects were extracted and the reference spectrum database was established as the basis for the subsequent classification of ground objects. Finally, aiming at the existing hyperspectral image recognition and classification methods, The spectral angle matching method (SAM) and the linear spectral decomposition method (LSUU, Linear Spectral mixing) were used to identify and classify the hyperspectral images before and after the optimal band selection and data fusion, respectively. The results show that the space-borne hyperspectral data can accurately identify the common types of urban features, and the recognition method is particularly important to the results. And the fusion processing of hyperspectral image can improve the precision and accuracy of the classification results to some extent. When the fusion image is identified by SAM method, The statistical error of the ground object area is only 11.61, but the LSU method is used to recognize the unfused image, the error will reach 65.63, and the result of the image is not clear, so it can not distinguish the distribution and aggregation form of each type of object.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：P237;TU984

【参考文献】