📄 中文摘要
高光谱图像捕捉了场景的大量高维光谱信息，使得标注成为一项耗时的任务，难以通过现成的统计方法进行处理。无监督学习聚类能够实现场景的自动分割，从而更快速地理解图像。通过在Wasserstein空间中进行字典学习来划分数据中的光谱信息已被证明是一种有效的无监督聚类方法。然而，这种方法需要平衡数据的光谱特征，导致类别模糊，并牺牲了对异常值和噪声的鲁棒性。该研究提出了一种改进的方法，利用非平衡Wasserstein重心来学习低维表示，从而提高无监督聚类的效果。
用于无监督高光谱图像聚类的非平衡最优传输字典学习
📄 English Summary
Unbalanced Optimal Transport Dictionary Learning for Unsupervised Hyperspectral Image Clustering
Hyperspectral images capture extensive high-dimensional spectral information about a scene, making labeling a labor-intensive task resistant to conventional statistical methods. Unsupervised clustering allows for automated segmentation of the scene, facilitating a quicker understanding of the image. Partitioning the spectral information through dictionary learning in Wasserstein space has proven effective for unsupervised clustering. However, this method requires balancing the spectral profiles of the data, which can blur classes and reduce robustness to outliers and noise. The study proposes an improved approach by utilizing unbalanced Wasserstein barycenters to learn low-dimensional representations, thereby enhancing the effectiveness of unsupervised clustering.