基于经验模分解高光谱图像数据尺度变换的光谱保真性分析

经验模分解是具有自适应性特点的尺度变换方法。高光谱图像数据近乎连续的光谱是开展信息提取的重要信息来源,然而尺度变换会使光谱发生变化。因此,分析EMD尺度变换后高光谱图像数据的光谱保真性具有重要意义。应用CHRIS高光谱图像数据,使用光谱相关系数、光谱偏差、光谱相对偏差和光谱角等评价指标,开展EMD升尺度图像及其典型地物的光谱保真性实验,并将EMD与Mallat小波变换光谱保真性比较。实验结果得出:11~10级EMD尺度变换后图像整体光谱保真性都较好,相关系数均在0.979以上,偏差小于55,相对偏差小于0.036,光谱角在0.041以内;2图像光谱保真性随EMD尺度变换次数增加而略有降低,且前4级变换光谱失真相对明显,后续降幅微弱;31~10级尺度变换后7种湿地典型地物的光谱保真性都较好,其中芦苇和河流的光谱保真程度较突出,养殖水面的相对不理想;4EMD与小波变换光谱保真性比较,随着变换次数的增加EMD表现出相对稳定且较小的光谱失真。

Spectral Fidelity Analysis of Scaling Transformation of Hyperspectral Remote Sensing Image based on Empirical Mode Decomposition

Empirical Mode Decomposition (EMD) is a method of image up\|scaling with adaptive characteristic.Almostly continuous spectrum of hyperspectral image data is the basic of information extraction,but the scaling transform can lead to spectal distort.Thus it is important to analyse the spectral fidelity of hyperspectral image after EMD up\|scaling transform.CHRIS hyperspectral remote sensing image is taken as experiment data.The Correction Coefficient(CC),Error,Relative Error(RE) and Spectral Angle Mapper(SAM) are used to evaluate the spectral fidelity of image and the typical ground objects.The results of EMD and Wavelat transform are compared.The results show that:①The level 1\|10 up\|scaling images have well spectral fidelity,average spectral correlation coefficient is more than 0.979,error is less than 55,reletive error is less than 0.036,and spectral angle mapper is within 0.041.②With increaing of up\|scaling,the transform images spectral fidelity is slowly declining,and the spectral distortion of former 4 levels is more obvious than the others.③The spectral fidelity of seven typical groud objects which is chosed from image are well.Phragmites and River spectral are preserved best,but the Aquaculture water specral matching is not good.④Compared with Wavelet trandform on image spectral fidelity,the method of EMD is raletivly stability and has less spectral distortion with up\|scaling increaing.