CNN Based Hyperspectral Image Classification Enhanced by Dimensionality Reduction Techniques

Abstract

Hyperspectral imaging (HSI) is a powerful technique for capturing high-density 3D images across multiple spectral bands. HSI is used for various tasks across different domains and several algorithms with diverse approaches have been applied to process HSI data. However, these algorithms often face significant challenges due to high computational costs associated with complex dimensionality. In this study, we adopt two strategic approaches to implement a novel method for HSI classification. The dimensionality reduction using metaheuristic optimization is implemented based on the initial solution derived by Principal Component Analysis (PCA). The Metaheuristic search algorithm- Tabu Search enhanced by the Naïve Bayes (TSNB) fitness method is used for the dimensionality reduction process.  This research aims to apply a Convolutional Neural Network (CNN) with dimensionality reduction techniques to simplify the classification process. We utilize our approach to demonstrate successful results in the HSI classification process. Our method shows promising results on the Indian Pine dataset compared to four other leading band selection techniques.