Band Selection Approach based on Modified Cuckoo Search Algorithm for Hyperspectral Image

The Band Selection refers to the procedure of selecting the most relevant bands in the Hyperspectral Image (HSI). The HSI consists of important number of adjacent dense spectral bands which continuously have greater redundancy and high correlation that results into “curse of dimensionality” in an analysis of HSI. Hence, the efficient band selection approaches are significant for the reduction of HSI dimensionality without the any loss of spectral data involved in it. In this research, the new selection method based on the Modified Cuckoo Search Optimization (CSO) is proposed for the reduction of data dimensions of HSI. The performance of the proposed method is evaluated by utilizing the HSI dataset. The MCSO examines the various land-cover classes as well as stability of the selected features among two scenes named leading to discriminative as well as domain-invariant features of the subset of selected features. The performance of the proposed MCSO method attains better results and it achieves the Overall Accuracy (OA) of 90.67%, Average Accuracy (AA) of 75.89% and Kappa coefficient (k) of 0.897 respectively when compared to the existing methods named Binary Particle Swarm Optimization with Local Band Index (LBI-BPSO), β-Whale Optimization Algorithm (β-WOA) and Modified Hybrid Rice Optimization (MHRO).