Crossover Dimensionality Reduction and Multi-Branch Deep Learning for Enhanced Hyperspectral Image Classification

The amalgamation of deep learning techniques has led to significant advancements in hyperspectral image (HSI) classification. However, the innate 3D cube structure of HSIs poses a distinctive challenge, demanding a novel strategy for efficient spectral data utilization in classification. This study centers on HSI classification using a recently validated deep-learning approach. Challenges in HSI classification include dimensionality, data redundancy, and computational costs, while CNN-based methods dominate due to architectural constraints. To overcome these hurdles, we present a pioneering "Crossover Dimensionality Reduction and Multi-branch Deep Learning" (CDMD) model for hyperspectral image classification. This approach involves a multi-branch deep learning model combined with Factor Analysis and MNF for crossover feature extraction, with optimal features selected from each technique. Experimental results underscore the CDMD model's superiority over existing methods, underscoring its potential for enhancing HSI classification outcomes. Notably, the CDMD model delivers outstanding performance on benchmark datasets — Salinas Scene (SC), and Kennedy Space Center (KSC) — achieving impressive overall accuracy rates of 99.35% and 99.18% using just 5% of the training data.