经验模态分解在高光谱遥感数据处理中的应用

发布时间：2018-05-01 00:26

本文选题：经验模态分解(EMD) + 特征模态函数(IMF)　；参考：《成都理工大学》2017年硕士论文

【摘要】：高光谱遥感具有高的光谱分辨率,能为像元提供几乎连续的波谱曲线,高光谱遥感具备反演地物细节的能力。高光谱数据是复杂的非线性非平稳信号,经验模态分解(Empirical Mode Decomposition,简称EMD)是一种新的自适应时频分析方法,经EMD分解后的各个特征模态函数能突出原始信号的局部特征,从而更加方便地对非线性非平稳信号进行处理与分析。因此,本文尝试将EMD方法应用于复杂的高光谱数据中。本文主要工作有:(1)仔细探究常见时频分析方法及其局限性,深入研究经验模态分解的基本原理以及Huang所提出的创新性概念--特征模态函数,对EMD算法的分解过程进行了详尽的论述,以及该算法自身所具备的优越性能,指出EMD分解过程中存在的问题以及针对此提出的研究方向。(2)在高光谱数据的获取过程中,由于受到各种因素的影响,产生大量噪声,从而影响数据的最终分析结果。基于此,根据随机噪声的特性,结合高光谱数据的变化特征,提出了一种基于自相关函数特性的EMD去噪方法。通过仿真实验结果表明,含噪信号经EMD分解后,噪声主要集中在高频IMF分量中,将对应含噪IMF分量进行滤波后,与剩余IMF分量进行重构,噪声与信号进行有效地分离,从而实现对高光谱数据滤波降噪的目的,通过比较,此方法优于小波去噪。同时,将该去噪方法应用于野外实测的岩心高光谱数据,结果显示,去噪效果良好。(3)由于高光谱数据的大数据量以及岩心高光谱信号具有多变性的特点,利用传统的SAM、SCA等方法去识别高光谱数据特征并不适用。将信号进行EMD分解,使信号分析真正实现了时频局部化。经EMD分解后的各个固有模态函数突出了原始信号的局部特征,以局部反映总体。本文尝试将EMD应用在高光谱数据的特征提取中,从而寻找一种能够方便对近似光谱识别的方法,并确保此方法是有效可行的。以黄铜矿与黄铁矿为实例,分解后,得到局部特征,经过对比分析特征模态函数,实现了对黄铜矿与黄铁矿的识别。
[Abstract]:Hyperspectral remote sensing has high spectral resolution and can provide almost continuous spectral curves for pixels. Hyperspectral remote sensing is capable of retrieving the details of ground objects. Hyperspectral data is a complex nonlinear non-stationary signal. Empirical Mode decomposition (EMD) is a new adaptive time-frequency analysis method. Each characteristic modal function after EMD decomposition can highlight the local characteristics of the original signal. Therefore, it is more convenient to process and analyze nonlinear nonstationary signals. Therefore, this paper attempts to apply EMD method to complex hyperspectral data. The main work of this paper is to explore the common time-frequency analysis methods and their limitations, to study the basic principle of empirical mode decomposition and the innovative concept proposed by Huang, which is called eigenmode function. In this paper, the decomposition process of EMD algorithm is discussed in detail, and the superior performance of the algorithm itself is discussed. It is pointed out that the problems existing in the process of EMD decomposition and the research direction. Because of the influence of various factors, a great deal of noise is produced, which affects the final analysis result of the data. Based on this, according to the characteristics of random noise and the variation of hyperspectral data, a EMD denoising method based on autocorrelation function is proposed. The simulation results show that the noise is mainly concentrated in the high frequency IMF component after the noise signal is decomposed by EMD. After filtering the corresponding noisy IMF component, the noise is reconstructed from the remaining IMF component, and the noise is effectively separated from the signal. In order to achieve the purpose of noise reduction of hyperspectral data filtering, this method is better than wavelet de-noising by comparison. At the same time, the method is applied to the field measured core hyperspectral data. The results show that the denoising effect is good. Traditional methods such as SAMU SCA are not suitable to identify hyperspectral data features. The signal is decomposed by EMD, and the time-frequency localization is realized in the signal analysis. Each inherent mode function after EMD decomposition highlights the local characteristics of the original signal and reflects the population locally. This paper attempts to apply EMD to feature extraction of hyperspectral data in order to find a method that can easily identify approximate spectra and ensure that this method is effective and feasible. Taking chalcopyrite and pyrite as examples, the local features are obtained after decomposition, and the identification of chalcopyrite and pyrite is realized by comparing and analyzing the characteristic modal functions.
【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】：TP751

【参考文献】