Hyperspectral Image Classification via Information Theoretic Dimension Reduction

Hyperspectral images (HSIs) are one of the most successfully used tools for precisely and potentially detecting key ground surfaces, vegetation, minerals. HSIs contain a large amount information about scene; therefore, object classification becomes difficult task such high-dimensional HSI data cube. Additionally, HSI’s spectral bands exhibit high correlation, creates dimensionality issues as well. Dimensionality reduction is, crucial step in pipeline. In order to identify pertinent subset features effective classification, this study proposes dimension method that combines feature extraction selection. particular, we exploited widely denoising minimum noise fraction (MNF) an theoretic-based strategy, cross-cumulative residual entropy (CCRE), Using normalized CCRE, redundancy maximum relevance (mRMR)-driven selection criteria were enhance quality selected feature. To assess effectiveness extracted features’ subsets, kernel support vector machine (KSVM) classifier was applied three publicly available HSIs. The experimental findings manifest discernible improvement accuracy qualities features. Specifically, proposed outperforms traditional methods investigated, with overall accuracies on Indian Pines, Washington DC Mall, Pavia University 97.44%, 99.71%, 98.35%, respectively.