基于分组三维离散余弦变换字典的植物高光谱数据去噪方法  

Denoising Method for Plant Hyperspectral Data Based on Grouped 3D Discrete Cosine Transform Dictionary

作　　者：徐平[1] 肖冲[1] 张竞成[1] 薛凌云[1] 

机构地区：[1]杭州电子科技大学生命信息与仪器工程学院,浙江杭州310018

出　　处：《光学学报》2017年第6期321-330,共10页Acta Optica Sinica

基　　金：国家自然科学基金(41671415);国家自然科学基金青年基金(61205200);国家重大科研仪器研制项目(61427808);浙江省公益技术应用研究项目(2016C32087)

摘　　要：针对植物高光谱图像各波段噪声强度不同,以及空间域和谱域均存在噪声污染的问题,提出了一种基于分组三维(3D)离散余弦变换(DCT)字典的稀疏表示去噪方法。首先分析了植物光谱特征,根据谱间相关性对波段进行分组;然后采用边缘块剔除的局部均值标准差法对高光谱图像进行噪声标准差估计,为去噪算法提供参考阈值;最后构建三维DCT字典的稀疏表示去噪方法,对植物高光谱图像进行去噪。实验结果表明,与原始数据和二维DCT字典去噪方法相比,谱域噪声评估中平均信噪比分别提高18.2dB和9.2dB。因此,该方法不仅具有较好的空间域去噪能力,也有较好的谱域去噪能力。In order to solve the problems that noise intensity of each band for plant hyperspectral image is different and noise exists in both spatial and spectral domains,a sparse representation denoising method is proposed based on the grouped three-dimensional（3D）discrete cosine transform（DCT）dictionary.Firstly,the spectral characteristics of the plants are analyzed and the bands are grouped according to the spectral correlation.Secondly,local mean standard deviation of eliminating edges is used to estimate the noise standard deviation of hyperspectral images,which provides the reference threshold for denoising algorithm.Finally,a sparse representation denoising method based on 3D DCT dictionary is constructed for denoising plant hyperspectral images.Experimental results show that,comparing with the original data and the denoising method of two-dimensional（2D）DCT dictionary,the average signal-to-noise ratios of the noise evaluation by the proposed method are improved by 18.2 dB and 9.2 dB in the spectral domain.Therefore,the proposed method can denoise not only in the spatial domain but also in the spectral domain.

关 键 词：光谱学 高光谱数据 植物光谱特征 三维离散余弦变换字典 二维离散余弦变换字典 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]