Multifractal based hyperion hyperspectral data mining

Traditional methods effectively used for processing multispectral data is usually limited to deal with hyperspectral images, which are characterized by massive data and higher spectral resolution. In this paper, multifractal is introduced to mining hyperspectral data by analyzing the holistic feature of spectral curve, and blanket method is adopted to compute fractal dimension (fractal signature) of each spectral curve within Hyperion image. The experimental result shows the advantage of fractal signature over original spectral to identify the objects, and the down fractal signature may be more effective to discriminate objects than up fractal signature for the Hyperion data. The experimental result also indicates that the initial scale affects the fractal signature value, and the number of bands affects both fractal signature value and feature scale. The primary result implies that the multifractal based method may be a feasible approach to mining hyperspectral data.