Optimal hyperspectral band selection using robust multi-verse optimization algorithm

Hyperspectral Images (HSI) generally have high spectral resolution when compared with multispectral and panchromatic images. On the other hand, hyperspectral images are redundant in nature and to prevent this band selection technique is adopted which is efficient to decrease the hyperspectral data size. Multi-Verse Optimizer (MVO) is a novel nature-inspired metaheuristic algorithm, and it is based on the theory of multiverse in astrophysics. MVO is used to select optimal features and the effectiveness is proved through selective bands verification by classification algorithms such as: Random Forest (RF), and Support Vector Machine (SVM). It was observed from the literature that the existing works on hyperspectral band selection lacks in convergence rate and optimal fitness value. The proposed work aims to develop an efficient algorithm for optimal band selection for hyperspectral image data and it can be further used for agricultural applications. As a pilot study, the MVO algorithm is applied on two available hyperspectral image datasets such as: Pavia university and Indian Pines. To verify the results, MVO is compared with Particle Swarm Optimization (PSO) and Hybrid PSO-MVO (HPSO-MVO) and Sparrow Search Algorithm (SSA). Fusion of PSO-MVO algorithm was attempted in this work and the effectiveness of hyperspectral band selection was found to be improved in terms of its execution time and convergence analysis. From the experimental results it was observed that the proposed algorithm gives the average accuracy of 92.5% for MVO, 90% for PSO and 91% for HPSO -MVO respectively using SVM. In addition, RF classifier is also performed which gives the average accuracy of 89.5% for MVO, 87.5% for PSO, 89% for HPSO - MVO and 88% for SSA respectively.