高光谱影像的二维EMD分解权系数消噪方法  

Hyperspectral image denoising based on bidimensional empirical mode decomposition combined with weight coefficient method

作　　者：厉祥[1] 王文波[1] 

机构地区：[1]武汉科技大学理学院,武汉430065

出　　处：《中国科技论文》2013年第7期617-620,共4页China Sciencepaper

基　　金：国家自然科学基金资助项目(41071270;11201354);武汉市晨光计划资助项目(201150431096)

摘　　要：高光谱遥感影像不仅带有地物的二维空间数据,而且带有地物的一维光谱数据,其影像的光谱数据之间具有很强的相关性。基于高光谱影像图谱合一和谱间信息相关性的特点,利用二维经验模态分解(BEMD)对高光谱影像进行消噪。首先通过BEMD将高光谱影像的各波段数据分解为一组固有模态函数;然后根据不同噪声强度的波段间光谱信息的对应关系,计算各波段的权系数值,对小噪声波段数据的固有模态函数系数进行加权求和,利用所求出的系数值替换强噪声波段的固有模态函数系数;最后对处理后的固有模态函数系数进行累加重构得到消噪后高光谱影像。利用高光谱影像进行了实验分析,实验结果表明,与小波消噪方法相比,高光谱影像经本文方法消噪后视觉效果更好,且具有更高的信噪比,在有效去除影像噪声的同时,可以更好地保留有用细节信息。The Hyperspectral imageries contain both the two-dimensional and one-dimensional spectral information data, having the characteristics of the spectrum as unity. The spectral information own the strong correlation. For these characteristics of hy- perspectral images, this paper presents a two-dimensional empirical mode decomposition based hyperspectral image noise reduc- tion method. Firstly, the method uses two-dimensional empirical mode decomposition for each band hyperspectral images were decomposed to get different scales of intrinsic mode functions. Then, according to including noise larger band and small noise band of the spectrum corresponding relation between the weight coefficient of value, to contain the small noise band high frequen- cy intrinsic mode function coefficient weighted sum, after using the weighted coefficient values instead of containing noise larger band high frequency intrinsic mode function coefficient. Finally, the denoised hyperspectral images are reconstructed by inverse two dimensional empirical mode decomposition. Experiments show that The method can effective denoising of hyperspectral im- age, at the same time also can well retain the image details. Compared with classic wavelet denoising method, using the method of denoising image has higher peak signal to noise ratio and better visual effect.

关 键 词：高光谱影像 二维经验模态分解 加权系数法 消噪 

分 类 号：P237[天文地球—摄影测量与遥感]