CFFT最优信噪比的星载高光谱影像噪声抑制方法

Hyper-spectral remote sensing image noise reducing by CFFT based on optimal SNR

高光谱遥感影像独特的"图谱"特征为定量遥感等遥感监测提供了重要的技术支撑,但由于大气、地形材料等因素影响和限制,易受到各种噪声污染。基于光谱学领域的光谱统处理理论,结合信息时域分析领域的快速傅里叶变换,实现频域内噪声与光谱有效信息的分离,同时在最优信噪比约束条件下选择最优滤波阈值,设计噪声滤波器,实现了卫星高光谱遥感影像的噪声滤除。以中国北京城区和中国新疆谢米斯台上戈壁区的两景Hyperion高光谱遥感影像为例,进行了该滤波算法的定性和定量评价。结果表明:文中提出的高光谱遥感影像噪声滤波算法,有效地滤除了高光谱影像数据光谱维高斯白噪声和空间维的条带噪声、波段差噪声,提高了影像的清晰度,影像的信噪比提高5 dB以上,为后续定量分析等遥感监测研究提供了较高质量的数据保障。

Space-borne hyperspectral remote sensing imagery, supplying both spatial and spectral information for quantitative remote sensing monitoring, is easily affected by noise from atmosphere and terrain etc. Based on the spectral continuum removing and recovering, traditional fast Fourier transform （FFT） was extended to continuum fast Fourier transform（CFFT） to separate noise from target information in frequency domain （FD）. Then low-pass filter for reserving useful information was designed to eliminate noise, with its cut-off frequency selected self-adaptively by optimal signal-to-noise ratio （SNR）. Hyperion hyperspectral imageries of Beijing China and Xinjiang China were singled out for noise removal to validate the filtering ability of the continuum fast Fourier transform self-adapted by optimal signal-noise ratio （CFFTOSNR） method with qualitative description and quantificational indexs, including mean, variance,entropy, definition and SNR etc. The experimental result shows that CFFTOSNR do well in reducing the Gauss white noises in spectral domain, stripe and band-subtracting noise in spatial domain respectively, while the quantificational indexs of filtered imagery are all improved, with entropy of post-processed image obviously increased by 5 dB.