Subgrouping-Based NMF with Imbalanced Class Handling for Hyperspectral Image Classification

Abstract

The remote sensing industry is actively discussing the classification of hyperspectral images (HSIs). For the first time, the idea of subgrouping dimensionality is presented using a modified deep learning model, and this research presents a novel framework for dimensionality reduction in HSI classification as a result. In particular, our system uses the subgrouping model to extract many characteristics from a dataset and then apply a selection criterion. First, we performed data reduction and subgrouping by extracting the correlation matrix. After that, we resample the data and use it as input for a hyperspectral picture classification. In the proposed framework, we combine NMF on spectral dimensions with information-based feature selection and a wavelet-based 2D CNN on spatial dimensions to classify spectral-spatial data. Based on the experimental findings, it is clear that this framework delivers the most excellent classification accuracy compared to other approaches, including traditional classifiers like PCA and MNF-based deep learning methods.