I. Introduction
Hyperspectral images (HSIs) contain hundreds of spectral bands, along with abundant spatial information, captured by hyperspectral sensors. In contrast to traditional multi-spectral images, HSIs facilitate distinguishing differences among various land covers. This makes them valuable in applications such as object detection and recognition [1], environmental monitoring [2], and disaster prediction [3]. The significant increase of spectral bands contributes to the improvement of specific tasks but can induce drawbacks such as the Hughes effects [4]. The high-dimensional HSIs often have redundant and noisy bands, which affects the accuracy of applications and increases the costs of computation and storage. To mitigate these issues, dimensionality reduction is often employed as a preprocessing step for processing HSIs.