Spectral-Spatial hyperspectral image classification based on hybrid of archimedes optimization algorithm and genetic algorithm with bitonic filter

Abstract

Hyperspectral imaging (HSI) is widely utilized for its ability to capture detailed spectral information about objects’ physical and chemical properties. However, the high dimensionality of HSI data introduces challenges, including redundant, noisy, and highly correlated spectral bands, which can degrade classification performance. Effective feature selection is essential to mitigate these issues and address the Hughes phenomenon by retaining only the most informative bands. Metaheuristic optimization techniques have been widely applied for band selection, but they often face limitations such as premature convergence and local optima entrapment. To overcome these challenges, this study proposes a hybrid optimization method, AOA-GA, which integrates the Archimedes Optimization Algorithm (AOA) with the Genetic Algorithm (GA). This hybrid approach achieves a superior exploration–exploitation balance, enabling efficient global optimization and reducing the risk of local optima. Additionally, spatial features are incorporated during classification using the Bitonic Filter, enhancing overall accuracy. The proposed method is evaluated using two classifiers, Random Forest and Support Vector Machine, on three benchmark hyperspectral datasets: Indian Pines, Pavia University, and Salinas. Experimental results demonstrate that AOA-GA significantly outperforms state-of-the-art methods, showcasing its robustness, efficiency, and superior classification performance across diverse HSI datasets.