Hiperspektral görüntüde boyut indirgeme yöntemleri

Hyperspectral Images has huge dimensions of data compared to single band or multispectral band images. This results from the fact that it contains hundreds of spectral bands with a high spectral resolution. Therefore, hyperspectral data processing, storing, and transmitting are critical issues to deal with. Additionally, it is a fact that required sample size for training a specific classification method increases exponentially with increasing number of spectral bands. In order to handle these problems, either the training data size has to be enlarged or dimensionality of hyperspectral images has to be reduced with some dimension reduction techniques. In this thesis, supervised and unsupervised dimension reduction methods are investigated, and some new methods are proposed. The proposed methods aim to reduce the dimensionality of the hyperspectral data before classification while preserving the classification accuracy as much as possible and to achieve reduced dimension with a low computational complexity.