An unsupervised band selection method for hyperspectral images using mutual information based dependence index
Abstract
Hyperspectral images are increasingly used in areas ranging from agriculture to astronomy. As these images contain features representing narrow contiguous bands, similarity between adjacent bands is high. This paper presents an un-supervised band selection method using a feature similarity index named mutual information based dependence index (MIDI). This index is capable of identifying non-linear relation between features. Experiments were done with Indian Pines and Botswana datasets and the proposed band selection method resulted in better classification performance in terms of overall accuracy and Kappa coefficient.
Authors: Namita Jain, Susmita Ghosh
Published in: IEEE International Geoscience and Remote Sensing Symposium (IGARSS) (2022)