Multivariate grey model based BEMD for hyperspectral classification

Bi-dimensional empirical mode decomposition (BEMD) has been one of the core activities in image processing. Unfortunately, this promising technique is sensitive to boundary effect. Here, a new technique based on multivariate grey model termed as GM(1, 3) is developed for boundary extension in BEMD. More specifically, pixel values and coordinates of the image are regarded as characteristic data series and relative data series of GM(1, 3), respectively. Therefore, the extended image is decomposed into several BIMFs and a residue. Eventually, the corresponding parts of the BIMFs as well as the final residue are extracted as the decomposition results of original image. The effectiveness of the proposed approach is tested on hyperspectral classification in which the generally acknowledged support vector machine (SVM) is adopted as classifier. Experimental results confirm the validity of the proposed method.