Simulated Annealing-Based Optimization for Band Selection in Hyperspectral Image Classification

Said Khelifa, Fatima Boukhatem, Leila Benaissa Kaddar


In this paper, a new optimizationbased framework for hyperspectral image classificationproblem is proposed. Band selection is a primordialstep in supervised/unsupervised hyperspectral imageclassification. It attempts to select an optimal subset ofspectral bands from the entire set of hyperspectral cube.This subset is considered as the relevant informativesubset of bands. The advantage of an efficient bandselection approach is to reduce the hughes phenomenonby removing irrelevant and redundant bands. In thisstudy, we propose a new objective function for theband selection problem by using Simulated Annealingas an optimization method. The proposed approach istested on three Hyperspectral Images largely used in theliterature. Experimental results show the performanceand efficiency of the proposed approach.


Optimization, band selection, classification, bagging, correlation, simulated annealing

Full Text: PDF