基于二维经验模态分解的高光谱影像去噪方法

Hyperspectral image denoising based on bidimensional empirical mode decomposition

高光谱遥感图像同时具有二维空间信息数据和一维光谱信息数据,具有图谱合一的特点且谱间信息具有强烈的相关性,针对高光谱图像的这些特点,提出一种基于二维经验模态分解的高光谱图像降噪方法。该方法利用二维经验模态分解对各波段的高光谱图像分别进行分解,得到不同尺度的固有模态函数;根据含噪声较大的波段和含噪声较小的波段的谱间对应关系计算权系数值,对含噪声较小波段的高频固有模态函数系数进行加权求和,利用加权后的系数值代替含噪声较大的波段的高频固有模态函数系数;利用去噪后的高频系数进行重构得到去噪后的高光谱图像。实验表明,该方法能够对高光谱影像进行有效去噪,同时亦能较好地保留图像细节信息,与经典的小波去噪方法相比,使用该方法去噪后的图像具有更高的峰值信噪比和更好的视觉效果。

The Hyperspectral images have the two-dimensional spatial information data and one- dimensional spec-tral information data, information between spectrum has strong correlation. Aiming at the characteristic of hyperspec-tral image, a hyperspectral imagery denoising method based on two dimensional empirical mode decomposition is pro-posed. At first, each band of the hyperspectral image is decomposed respectively by two dimensional empirical mode, and different scales of intrinsic mode functions are obtained. According to spectral correlation between the bands con- taining seriously noise and the bands containing low noise, the weight values are calculated, The high frequency intrin-sic mode functions coefficients of the bands containing low noise are weighted and summed. The original intrinsic mode functions coefficients of the band containing seriously noise are replaced by the sum of weighted high intrinsic mode functions coefficients of the band which contains weak noise. Finally,the denoised hyperspectral images are a- chieved by inverse two dimensional empirical mode decomposition. The experimental results show that the proposed method can remove the noise of hyperspectral image effectively and keep the detail well. Comparing with the classical wavelet denosing method, the denoised images by the proposed method have a higher SNR and better visual effects.