Hyperspectral unmixing (HU) is a very useful and increasingly popular preprocessing step for a wide range of hyperspectral image analyses. However, the HU study has been constrained by the limited hyperspectral images (especially the ones with ground truths). In this paper, we propose a method to label the hyperspectral images. Besides, we summarize the most commonly used hyperspectral images for the HU study, including their scenes, hyperspectral sensors, resolutions, etc. Although those images and their ground truths are widely used in tens of HU papers, it is worth emphasizing that most of them are, unfortunately, not available on the web. Such case hinders the progress pace of new HU methods. Thus, we give a thorough summarization of 18 real images and their ground truths, including the endmembers and abundance maps, and provide them on the web. To the best of our knowledge, this is the first paper that summarizes and provides the real hyperspectral images and their ground truths for the HU study.
View on arXiv