The influence of kernel principle componets based feature extraction on hyperspectral image classification accuracy

Image data which belonging to many narrow wave bands are acquired with hyperspectral remote sensors and as a result a decomposition with respect to wave length is achieved. Because the acquired data amount is large, feature extraction is an important research subject. In this paper, the effect of the recently proposed kernel principle component (KPC) based hyperspectral feature extraction approach on classification accuracy is investigated. While the approach is shown in the literature to improve classification accuracy when used wit linear classifiers, it is shown in this paper that the approach cannot reach the performance of non-linear classifiers.