CASIA OpenIR  > 毕业生  > 博士学位论文
高光谱图像降噪算法研究
Alternative TitleResearch on Denoising of Hyperspectral Imagery
陈绍林
Subtype工学博士
Thesis Advisor彭思龙
2012-08-09
Degree Grantor中国科学院研究生院
Place of Conferral中国科学院自动化研究所
Degree Discipline模式识别与智能系统
Keyword高光谱图像 降噪 多尺度平滑 光谱平滑性和连续性 混合先验 空间自适应的光谱先验 混合矢量全变分 Hyperspectral Images Denoising Multiscale Smoothing Spectral Continuity And Smoothness Mixing Prior Spatially Adaptive Spectral Prior Hybrid Vectorial Total Variation
Abstract高光谱成像技术是目前遥感技术发展的一个前沿技术,已被成功应用于许多领域。高光谱成像技术一个突出的优势是,在采集反应地物空间与几何特性的二维图像的同时,获取地物连续的光谱信息。 高光谱遥感图像的``图谱合一''的特点给遥感技术的应用带来了巨大的变革。一方面,深化了传统遥感图像的应用,如利用地物光谱属性,可以对地物间微小差别进行精细识别;另一方面,拓展了遥感应用的分析手段,将传统遥感侧重于定性分析推进到定量分析中,如高光谱解混(Umixing)技术在确定端元(Endmember)光谱的同时,还可以获得每种端元在像元内含有的比例。 高光谱成像技术的应用使得基于图像和光谱的遥感技术的应用逐渐朝更加精细准确的方向发展,对图像质量也提出了更高的要求。 在高光谱成像的过程中,受到采集环境、传感器噪声和其它一些不确定因素的干扰,采集的原始数据中往往含有各种噪声。噪声的存在降低了图像质量,抑制了有用信息,影响信息提取的精度,有时甚至会导致完全错误的结论。因此对星载或机载获取的高光谱图像进行预处理,降低噪声,为后续各种应用提供更高质量的图像,是获取 高光谱图像后的首要工作。 与传统的宽波段遥感相比,成像光谱仪的成像质量更易受到环境的干扰,且其噪声的来源和特点也更为复杂。高光谱图像可以 看作是包含两个空间维和一个光谱维三维张量,其特点有别于传统的多光谱图像。由于传统的灰度图像和多光谱图像的 降噪算法已经无法满足高光谱图像降噪的需求,因此根据高光谱数据的特点研究新的降噪算法已成为高光谱图像处理领域的一个新热点。 本文在分析影响光谱仪成像质量的各种因素、高光谱图像噪声的来源及特点的基础上,针对高光谱图像的特点,综合考虑了信号在空间域和光谱域的不同属性,提出几种降低高光谱图像随机噪声的策略。例如,一方面,对空间域和光谱域信号可以采用的不同先验,或是采用不同的正则化参数;另一方面,利用光谱属性随空间变化的特性以及噪声统计量随波长变化特点。本文的创新成果主要有以下三点: 1. 提出了一种基于多尺度平滑的高光谱图像降噪框架。该框架克服了传统的光谱域降噪无法利用空间邻域信息,以及 二维的空间域降噪无法利用相邻波段间较强的相关性的缺陷。该框架针对高光谱图像``图谱合一''的特点,充分利用了空间图像在小波变换域能量的集中性和系数的稀疏性,以及像元光谱具有较强的连续性和平滑性的特点。首先对每个波段的光谱图像单独进行小波变换;然后在小波变换域内,沿着光谱轴方向利用三次样条平滑对含有噪声的信号进行降噪;最后对降噪后的系数信号进行小波逆变换。为了适应高光谱图像噪声强度随波长变化的特点,估计各个波段噪声方差来调节局部的平滑程度。 2. 基于最大后验(Maximum a Posteriori)概率模型,提出了一种空间域和光谱域混合的先验降噪算法。针对空间域和光谱域信号表现出的不同性质,在建模的过程中, 对空间域和光谱域信号采用不同的先验。并且针对不同物质的光谱具有不同的平滑性,提出了一种空间自适应加权的光谱先验。 该先验定义为平滑性和不连续先验的加权和...
Other AbstractHyperspectral imaging is one of the cutting edge technology in the remote sensing field and it has many successful applications. One of prominent advantages of hyperspectral imaging technology is it can acquire the continuous spectral information of materials while collecting the geometric information of ground objects. This characteristic of hyperspectral images (HSIs) has brought tremendous changes for the applications of the remote sensing technology. On one hand, it deepens the applications of the remote sensing technology, for example, by utilizing spectral features of materials, we can distinguish two materials with very small difference; on the other hand, it provides more means for the applications of remote sensing and make the quantitative analysis of objects become possible, for instance, the unmixing techniques of HSIs not only can determine the spectra of endmembers, but also can determine the fraction of each endmember in each pixel. The development of hyperspectral imaging technology makes the applications of remote sensing technology based on the images and spectra toward a more refined and accurate direction, it also puts forward higher requirements on image quality. The process of hyperspectral imaging is generally interfered by the acquisition environment, the sensor noise and other uncertain factors, thus the acquired raw data often contain a variety of noise. The noise will degrade the image and has a negative effect on information extraction. To reduce the noise of the HSIs acquired by spaceborne or airborne imaging spectroscopy and enhance the qualities of HSIs is the pre-requisite task for the processing of HSIs. Compared with the traditional wide-band remote sensing, the image qualities of imaging spectroscopy are more vulnerable to inferences from capturing environment, such as the atmospherical absorption and scattering. The noise sources of HSIs and their characteristics are also more complex. The HSI is a 3D tensor which includes two spatial dimensions and one spectral dimension, its characteristics are different from the traditional multi-spectral images. Thus, denoising approaches for the traditional images are not suitable to HSIs and we should investigate new denoising methods according to the characteristics of HSIs. Based on analyzing the factors influencing the imaging quality of spectral imaging, the noise sources and noise characteristics of HSIs, we propose several denoising schemes for HSIs in this thesis. Dur...
shelfnumXWLW1819
Other Identifier200818014628031
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/6481
Collection毕业生_博士学位论文
Recommended Citation
GB/T 7714
陈绍林. 高光谱图像降噪算法研究[D]. 中国科学院自动化研究所. 中国科学院研究生院,2012.
Files in This Item:
File Name/Size DocType Version Access License
CASIA_20081801462803(10358KB) 暂不开放CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[陈绍林]'s Articles
Baidu academic
Similar articles in Baidu academic
[陈绍林]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[陈绍林]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.