A New Approach to Compute Pixel Vector-Based Morphological Profile for Classification of Hyperspectral Image

Analysing hyperspectral images using supervised classification is challenging, primarily because of the higher spectral dimensions and the presence of rich and complex spatial information. This article proposes a strategy for pixel vector ordering to compute multiband morphological operations such as dilation, and erosion and to analyse features using multiband morphological profiles before classification. The proposed vector ordering strategy assigns weights to each band, computed using dilation distance, for multiband morphological operations. The features obtained from computing a multiband morphological profile using the proposed vector ordering strategy are used consequently for the classification of the data using Support Vector Machine (SVM). For further verification, the method is experimented on well-known hyperspectral image data “Pavia University” and “Indian Pines”. The experimental results show that classification using the multiband morphological profiles using the proposed vector order strategy can surpass the classification accuracy obtained from multiband profiles using other vector ordering strategies when extracting pivotal features for classification.