科研之友
微信
新浪微博
Facebook
分享链接
摘要
Planetary space experiments collect large volumes of data whose scientific content requires understanding. Marzo et al. (2006) presented an unsupervised cluster analysis scheme that is able to reduce a spectral data set to a few clusters, allowing for more focused and rapid evaluation of their scientific meaning. Here, we extend the original approach to account for the measurement uncertainty and build a classification scheme. We apply the clustering technique to the ASTER and RELAB libraries of visible and infrared spectral reflectance. These spectral libraries are documented, allowing assignment of a label to each spectrum reflecting its physical and chemical properties. We assess the ability of the original and extended approaches to identify natural clusters of the library spectra and estimate associated uncertainties of the results. We evaluate the scientific meaning of the derived clusters based on the labels contained within each cluster. Once the cluster meanings are defined, we test our classification scheme using a training-testing approach and evaluate the accuracy of assigning the unknown spectra to the correct cluster.
- 出版日期2009-8-11
