Enhanced hyperspectral image analysis via 2D-3D CNN fusion and hybrid moth-flame optimization for optimal band selection in remote sensing

Classification of hyperspectral images is a promptly emerging area related to remote sensing. However, encounters arise in the form the large volume of data, unremitting acquisition processes, and correlations among spectral bands, which obfuscate the extraction of informative features. To tackle these issues, this paper presents a progressive framework for hyperspectral image analysis and classification. This framework combines 2D-3D Convolutional Neural Networks (CNN) with a Hybrid Moth-Flame Optimization (MFO) technique for optimal band selection. By focusing on the most relevant spectral bands, the proposed framework effectively handles the challenges associated and improving classification accuracy. The key novelty of this study lies in the amalgamation of deep learning with metaheuristic optimization, offering a robust approach to feature extraction and classification. Comprehensive experiments were carried out on three benchmark hyperspectral datasets—Indian Pines, Pavia University, and Salinas. The resulted quantitative metrics like overall accuracy rates between 89% and 97%, with Kappa coefficients ranging from 0.89 to 0.96. These findings emphasize the advantages of the proposed method over existing state-of-the-art techniques, demonstrating improvements in both classification accuracy and computational efficiency.