高光谱遥感图像的小波去噪方法

Hyperspectral Imagery Denoising Method Based on Wavelets

高光谱遥感图像是由二维空间信息和一维光谱信息组成的三维数据。普通的去噪方式通常是分别对空间信息或光谱信息进行去噪，其主要缺点是忽视了高光谱图像强烈的谱间相关性和图谱合一的特点。针对这些特点，文章提出一种基于小波变换的高光谱遥感图像去噪方法。该方法对各波段高光谱图像逐一进行二维小波变换，根据含噪声大的波段与噪声小的波段的波长关系，对小噪声波段的高频系数加权求和，代替噪声大的波段的高频系数，通过小波逆变换得到去噪后的重构图像。该方法运算速度快，能有效地降低噪声。对机载可见红外成像光谱仪数据（AVIRIS）实验表明，与经典的BayesShrink图像去噪方法相比，方法重构图像的信噪比（SNR）高出3.8～10.6 db，节省运算时间一半以上。

To take advantage of the intrinsic characteristic of hyperspectral imageries, a hyperspectral imagery denoising method based on wavelet transform is proposed in the present paper. At first, two dimensional wavelet transform is performed on hyperspectral images band by band to capture their profiles. Due to the significant spectral correlation between adjacent bands, their high frequency wavelet coefficients are similar as well. Then, according to the wavelength relationship among the bands, which contain noise with different variances, new high frequency wavelet coefficients of seriously noisy bands are computed by the sum of weighted high frequency wavelet coefficients of bands, which contain low variance noise, and their profiles destroyed by noise are recovered in this way. Finally, the denoised images are reconstructed through inverse wavelet transform. The proposed method runs fast and can remove the noise efficiently. It was tested on airborne visible/infrared imaging spectrometer data （AVIRIS） cubes. Experimental results show that the signal-to-noise-ratio （SNR） of the reconstructed images in our method is 3.8-10. 6 db higher than the that of the reconstructed images in the classical image denoising method, BayesShrink, and our method saves more than 50% computing time than BayesShrink method.