Hyperspectral multi-level image thresholding using qutrit genetic algorithm

Abstract

Hyperspectral images contain rich spectral information about the captured area. Exploiting the vast and redundant information, makes segmentation a difficult task. In this paper, a Qutrit Genetic Algorithm is proposed which exploits qutrit based chromosomes for optimization. Ternary quantum logic based selection and crossover operators are introduced in this paper. A new qutrit based mutation operator is also introduced to bring diversity in the off-springs. In the preprocessing stage two methods, called Interactive Information method and Band Selection Convolutional Neural Network are used for band selection. The modified Otsu Criterion and Masi entropy are employed as the fitness functions to obtain optimum thresholds. A quantum based disaster operation is applied to prevent the quantum population from getting stuck in local optima. The proposed algorithm is applied on the Salinas Dataset, the Pavia Centre Dataset and the Indian Pines dataset for experimental purpose. It is compared with classical Genetic Algorithm, Particle Swarm Optimization, Ant Colony Optimization, Gray Wolf Optimizer, Harris Hawk Optimization, Qubit Genetic Algorithm and Qubit Particle Swarm Optimization to establish its effectiveness. The peak signal-to-noise ratio and Sørensen-Dice Similarity Index are applied to the thresholded images to determine the segmentation accuracy. The segmented images obtained from the proposed method are also compared with those obtained by two supervised methods, viz., U-Net and Hybrid Spectral Convolutional Neural Network. In addition to this, a statistical superiority test, called the one-way ANOVA test, is also conducted to judge the efficacy of the proposed algorithm. Finally, the proposed algorithm is also tested on various real life images to establish its diversity and efficiency.