A new feature extraction scheme based on support optimization in Enhanced Multivariance Products Representation for Hyperspectral Imagery

In computational science, the preservation of characteristic data properties and the computational efficiency for processing practices amidst growing complexity is crucial. Hyperspectral Imaging, a domain of high-dimensional and dense datasets, requires effective preprocessing techniques to extract meaningful insights while also benefiting from the utilization of computational techniques. This work introduces a novel approach leveraging the Enhanced Multivariance Products Representation by iteratively optimizing its support vectors through the employment of the Alternating Direction Method of Multipliers, capitalizing on its decorrelation properties. The proposed method offers a feature extraction procedure for improving classification performance, by transforming the data while retaining its intrinsic characteristics. Experiments conducted on hyperspectral datasets showcase the efficacy of the proposed method in enhancing data representation for subsequent classification tasks. Through comprehensive analyses, the study highlights the achievement of the proposed technique in keeping essential data characteristics while mitigating noise and artefacts, thereby facilitating more accurate classification outcomes. Notably, the proposed method demonstrates its adaptability to varying dataset dimensions, underscoring its applicability in diverse Hyperspectral Imaging scenarios.