基于经验模态分解的高光谱遥感数据去噪方法

De-Nosing Method of Hyperspectral Data Based on EMD

经验模态分解(EMD)是一种新的时频分析方法,经EMD分解后的各个固有模态函数(IMF)突出了原始信号的局部特征,从而可以区分噪声和有用信号。基于此,结合高光谱遥感数据的光谱变化特征,提出了一种基于经验模态分解的高光谱遥感数据去噪方法。通过对理论数据的实验表明,数据中的噪声无论是高斯分布还是均匀分布,数据经EMD分解后,噪声都主要集中在前几个特定的IMF,对相应的IMF进行滤波处理后并与其他IMF分量进行重构就可得到去噪信号,与小波去噪结果相比较,这种方法效果更好。最后把该去噪方法应用于野外实测的油膜高光谱数据去噪,实验结果表明,该方法能准确、有效地去除高光谱遥感数据的噪声。

Empirical mode decomposition （EMD） is a relatively new time-frequency analysis method.IMFs were developed by EMD outstand local characteristics of the original signal,so they can discriminate the signals from the noise.An EMD based approach to hyperspectral data de-noising is proposed in the work.The simulative experiment with theoretical data shows that Gaussian-distributed noise or uniform-distributed noise inhered in the data is mainly concentrated in the first few specific IMFs.Therefore,the filtered noise-related IMFs together with the other IMFs can be used to restructure the denoised signal.Experimental result shows that the proposed method is more effective than wavelet-based method.Finally,this method is applied to denoise field oil-slick hyperspectral data and the results show that the method can denoise the inherent noise in hyperspectral data accurately and effectively.