A Comparative Study of Dimensionality Reduction Techniques for Satellite Image Analysis

An investigation into the use of dimensionality reduction on hyperspectral images and its effect on further classification is presented. The use of 5 different dimensionality reduction techniques along side 4 different classifiers has been done. The effect of the dimensionality reduction on the two datasets chosen, ROSIS City of Pavia and ROSIS University of Pavia, showed that there is merit to using dimensionality reduction on these datasets, and the results generated are comparable to those found in literature. Further investigation is also conducted on the use of a multiple classifier system, and the results generated by this method are better than any individual classifier on its own.