Mitigating Hughes Phenomenon: Improving Hyperspectral Imaging Classification Through Active Learning for Generalization Enhancement

In the domain of hyperspectral imaging classification (HIC), the challenge of limited labeled training samples, commonly referred to as the Hughes phenomenon, poses a significant obstacle. As hyperspectral datasets capture extensive spectral information, the Hughes phenomenon manifests when labeled samples are insufficient compared to the numerous spectral bands, resulting in reduced classification accuracy. This paper addresses the ill-posed conditions introduced by the Hughes phenomenon and explores how it adversely impacts classification algorithms in hyperspectral imaging (HI). Active learning (AL) emerges as a strategic solution to counteract the limitations posed by ill-posed conditions. By iteratively selecting and annotating the most important samples, AL allows for the augmentation of the training set without the need for an extensive collection of labeled data. This proactive approach mitigates the detrimental effects of limited labeled samples, enhancing the generalization performance of classification algorithms in HI.