2D2PCA-BASED HYPERSPECTRAL IMAGE CLASSIFICATION WITH UTILIZATION OF SPATIAL INFORMATION

Abstract

Classification of hyperspectral data is computationally complex and time consuming process due to the dimensionality of spectral signatures and high volume of data. In this study as a solution to this problem, an improved principal component analysis technique is proposed to extract features called as two directional-two dimensional principal component analysis (2D(2)PCA). For using 2D2PCA with hyperspectral images, each pixel is considered as a pixel set image with its surrounding neighbor pixels to utilize both spatial and spectral information. Exploitation of spectro-spatial information for feature extraction is more efficient and discriminative than using only one of them. In the comparative experiments classification accuracies are positively affected and increased by employment of both spatial and spectral features together.