Quantum fractional order Darwinian particle swarm optimization for hyperspectral multi-level image thresholding

Abstract

A Hyperspectral Image (HSI) is a data cube consisting of hundreds of spatial images. Each captured spatial band is an image at a particular wavelength. Thresholding of these images is itself a tedious task. Two procedures, viz., Qubit Fractional Order Particle Swarm Optimization and Qutrit Fractional Order Particle Swarm Optimization are proposed in this paper for HSI thresholding. The Improved Subspace Decomposition Algorithm, Principal Component Analysis, and a Band Selection Convolutional Neural Network are used in the preprocessing stage for band reduction or informative band selection. For optimal segmentation of the HSI, modified Otsu’s criterion, Masi entropy and Tsallis entropy are used. A new method for quantum disaster operation is implemented to prevent the algorithm from getting stuck into local optima. The implementations are carried out on three well known datasets viz., the Indian Pines, the Pavia University and the Xuzhou HYSPEX. The proposed methods are compared with state-of-the-art methods viz., Particle Swarm Optimization (PSO), Ant Colony Optimization, Darwinian Particle Swarm Optimization, Fractional Order Particle Swarm Optimization, Exponential Decay Weight PSO and Heterogeneous Comprehensive Learning PSO concerning the optimal thresholds, best fitness value, computational time, mean and standard deviation of fitness values. Furthermore, the performance of each method is validated with Peak signal-to-noise ratio and Sørensen–Dice Similarity Index. The Kruskal–Wallis test, a statistical significance test, is conducted to establish the superiority in favor of the proposed methods. The proposed algorithms are also implemented on some benchmark functions and real life images to establish their universality.