The multisperctral trait of hyperspectral RS is a new technology for RS image recognition and classification, on the other
hand, it is difficult to image processing owing to trait of data redundancy. This paper propose new method for
hyperspectral RS image classification. In order to reduce dimension, utilizing the hyperspectral RS's refined spectral
characteristic, we extract every pixel's spectral characteristic curve, and compute the fractal dimension of the curve. By
studying the relation between object and spectral characteristic curve and fractal dimension, the paper indicates that the
dilation fractal dimension is equal or close to same target wherever it locates, and different from different target. Then
based on every pixel's fractal dimension that interval is from 1 to 2, we stretch linearly the interval from 0 to 255, and
construct a new gray image. Lastly, we apply the approximate classing of rough set theory class to the new image, the result of classing is namely the result of hyperspectral RS image classification.
|