Comparative Assessment of Spectral Reduction Techniques and Machine Learning Classifiers for Hyperspectral Image Classification

High-dimensional hyperspectral images (HSI) offer detailed spectral information but pose challenges due to their complexity. This study addresses these challenges by comparatively analyzing four spectral reduction techniques: Principal Component Analysis (PCA), Independent Component Analysis (ICA), Least Absolute Shrinkage and Selection Operator (LASSO), and Random Forest. These techniques are forms of feature extraction and feature selection spectral reduction methods. Their impact on HSI classification was evaluated using various Machine Learning (ML) and Deep Learning (DL) classifiers on three benchmark datasets: Indian Pines, Pavia University, and Botswana. Performance metrics, including overall accuracy, kappa statistic, and confusion matrix analysis, were used for comparison. Results highlighted the robustness of the Random Forest classifier and the compatibility of Neural Networks classifier with feature extraction reduction methods, providing valuable insights into the effectiveness of spectral reduction techniques for improved HSI classification.