高光谱遥感混合像元端元提取研究及应用

发布时间：2018-06-19 08:00

本文选题：混合像元 + 端元提取　；参考：《中南大学》2013年硕士论文

【摘要】：高光谱遥感技术的核心依据是地物对电磁波的发射、吸收与反射特性。高光谱遥感就是最大限度地提高光谱分辨率,以光谱差异为核心依据实现地物精细分类的新兴遥感技术。正因为其光谱分辨率高、波段数多、数据量大的特点,高光谱遥感数据的空间分辨率往往不够高。混合像元普遍存在于高光谱遥感影像中,一定程度上阻碍了分类精度的提高。混合像元分解是进一步提高高光谱遥感分类精度必须面对的问题。混合像元分解分为端元提取和光谱解混两个过程,本文着重对端元提取进行了研究,并将提取的端元应用到光谱解混模型中,对光谱解混结果进行了对比分析。高光谱遥感影像数据中的像元在光谱空间中呈凸面几何分布,而端元就位于凸面几何体的顶点位置,这是端元提取的理论基础。本文介绍了现有的光谱混合模型和端元提取方法,并指出了每种模型方法的优缺点,并用PPI算法进行了端元提取,利用线性波谱分离技术进行了完整的光谱解混。本文以最大距离法为基础,针对该方法在提取初始端元上的不足,深入研究端元像元的光谱特性,将其转化为数学中的坐标进行分析,发现了端元具有的坐标特性,从而得出一种快速识别初始端元的方法,提取了五种矿物的端元波谱。用纯像元指数和波段分析工具对端元波谱进行了验证,并进行了光谱解混。同时,考虑了遥感影像在获取和处理过程中产生的误差,引入距离阈值概念,旨在提取与端元距离小于距离阈值的像元组成一组样本,求这些样本的平均光谱来代替原始端元。将这些平均光谱与USGS光谱库中对应矿物波谱进行匹配,证明了平均光谱的相似度较单一像元的端元有较大的提高,并进行了光谱解混。把三种方法提取的端元以及对应的光谱解混结果进行了对比分析,实验表明使用平均光谱能提高提取的端元的可靠性以及光谱解混的精度。
[Abstract]:The core of hyperspectral remote sensing technology is the emission, absorption and reflection characteristics of electromagnetic waves. Hyperspectral remote sensing is a new remote sensing technology which can maximize spectral resolution and realize fine classification of ground objects based on spectral difference. The spatial resolution of hyperspectral remote sensing data is often not high enough because of its high spectral resolution, large number of bands and large amount of data. Mixed pixels are widely used in hyperspectral remote sensing images, which to some extent hinder the improvement of classification accuracy. Mixed pixel decomposition is a problem that must be faced to improve the classification accuracy of hyperspectral remote sensing. The mixed pixel decomposition is divided into two processes: End-component extraction and spectral unmixing. In this paper, the End-component extraction is studied and applied to the spectral de-mixing model, and the results of spectral unmixing are compared and analyzed. The pixel in hyperspectral remote sensing image is geometric distribution of convex surface in spectral space, and the endmember is located at the vertex position of convex geometry, which is the theoretical basis of endmember extraction. In this paper, the existing spectral mixing models and end-component extraction methods are introduced, and the advantages and disadvantages of each method are pointed out. The end component extraction is carried out using PPI algorithm and the complete spectral unmixing is carried out by using linear spectral separation technique. In this paper, based on the maximum distance method, the spectral characteristics of the end element pixel are deeply studied in order to solve the shortcomings of the method in extracting the initial end element, and the coordinate characteristic of the end element is found out by transforming it into the coordinate in mathematics. Thus, a fast method for identifying the initial end components is obtained, and the end component spectra of five minerals are extracted. The end-element spectrum was verified by pure pixel exponent and band analysis tool, and the spectral unmixing was carried out. At the same time, taking into account the errors in the acquisition and processing of remote sensing images, the concept of distance threshold is introduced in order to extract a group of samples from pixels whose distance is less than the distance threshold, and to find out the average spectrum of these samples to replace the original endpoints. By matching these spectra with the corresponding mineral spectra in the USGS spectral library, it is proved that the similarity of the average spectra is much higher than that of the end elements of a single pixel, and the spectral unmixing is carried out. The results of the three methods are compared and analyzed. The experimental results show that the average spectrum can improve the reliability and the precision of spectral unmixing.
【学位授予单位】：中南大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：P237

【共引文献】