Hyperspectral images classification based on wavelet threshold denoising and empirical mode decomposition

In order to remove noise and achieve high accuracy classification of hyperspectral images, a high-accuracy hyperspectral images classification algorithm based on wavelet threshold denoising (WTD) and empirical modal decomposition (EMD) is presented. First, high-frequency noise in hyperspectral images is removed by wavelet threshold denoising. Second, the essential characteristics of hyperspectral images are extracted through the decomposition of hyperspectral images with EMD, and the residual with low-frequency noise is removed. Finally, the hyperspectral images are classified with SVM, which have been composed by the Intrinsic Modal Function (IMF) of hyperspectral images. Experimental results of the AVIRIS data indicate that the proposed approach not only improves the classification accuracy of hyperspectral images, but also reduces the number of support vectors and improves the speed of hyperspectral images classification.