Multistrategy Multiobjective Evolutionary Algorithm-Based Unsupervised Band Selection for Hyperspectral Image Classification

Hyperspectral images (HSI) contain abundant spectral information, but also exhibit significant redundancy. Therefore, selecting bands that possess high information content and low inter-band correlation is essential. Evolutionary algorithms (EAs) have demonstrated strong capabilities in addressing band selection (BS) problems. However, many existing methods rely on single-objective optimization, which limits their ability to jointly optimize multiple characteristics of the selected band subsets. To address this limitation, this study proposes an improved multi-objective evolutionary algorithm for band selection (IMOEABS). The proposed algorithm employs inter-class distance and information entropy as objective functions to identify bands with strong class discriminability and high information richness. It is built upon the dimension-wise special crowding distance (DSCD) non-dominated sorting mechanism, which penalizes small index differences between adjacent bands to suppress redundancy in the decision space. Furthermore, the generation strategy of offspring solutions is guided by the population entropy, enhancing the exploration efficiency of the algorithm.Experimental results on three benchmark datasets demonstrate that IMOEABS achieves superior performance compared to state-of-the-art BS methods, when evaluated using Support Vector Machine (SVM) and K-Nearest Neighbor (KNN) classifiers. These findings confirm the robustness and effectiveness of IMOEABS in hyperspectral band selection tasks.