No Detail Found!

Spectral unmixing with hyperspectral images: a multi-objective lion optimization approach

Author(s)
RAJA NATARAJAN Department of Information Technology, Bharathiar University, Coimbatore, India; (Computer Science), Tamil Nadu Agricultural University, Coimbatore, India
VADIVEL RAMASAMY Department of Information Technology, Bharathiar University, Coimbatore
Abstract
Spectral unmixing is an important phase in interpreting hyperspectral images to estimate the endmembers and their corresponding fractional abundances from the mixed pixels. Sparse unmixing approach is one among the promising spectral unmixing methods; however, noises impair hyperspectral images, indicating that the sparse unmixing model's performance getting degraded. This study uses a multi-objective lion optimization technique for spectral unmixing from hyperspectral satellite images to extract fractional abundances of valid endmembers from mixed pixels. To estimate the fractional abundances of each endmember, the proposed approach defines a multi-objective function which includes Euclidean and Frobenius norms and optimization variables that are associated with the abundance and mixing matrices. Based on the lion's social behaviors, the optimization function uses pride generation for finding solutions, mating for generating new solutions, territorial takeover and territorial defense to identify and remove worse solution by introducing a new optimum solution. The major objectives used in the objective functions are Spatial Neighbor, Sparsity, and Reconstruction Error. The proposed technique's performance is evaluated using Urban and Cuprite datasets with 06 and 12 endmembers respectively. In the investigations, the proposed methodology outperformed the Bilinear and Trilinear Multi-Objective Spectral Unmixing, Robust Collaborative Nonnegative Matrix Factorization, Pareto-Multi-Objective Spectral Unmixing, and Rider Optimization methods in estimating the enhanced endmembers and their corresponding fractional abundances with lower reconstruction error and root mean square error values of 119.11 and 0.0904, respectively.
Publication Details
Page(s) 2271-2288
DOI DOI not available
Published Journal: Journal of Theoretical and Applied Information Technology, Volume: 100, Issue: 7, Year: 2022

Request the full text from PASTIC

Keywords
Multi-objective Optimization, Hyperspectral Images, Endmembers, Spectral Unmixing
0

Citations

0

Downloads

24

Views