Accelerating Spectral Unmixing By Using Clustered Images
2017 24TH IEEE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING (ICIP)(2017)
摘要
We propose the usage of a clustering step before the unmixing of hyperspectral images This circumvents or at least mitigates the drawback of having a large amount of data that has to be processed by taking advantage of the large number of similar pixels and merging similar pixels into clusters. Afterwards, only the cluster centroids have to be unmixed instead of all image pixels. We call this meta-approach UNmixing of Clustered Image (UNCLI). It is especially useful for images with large more or less homogeneous regions. Another natural area of application are unmixing algorithms with spectral regularization because the calculation of the abundance map very often is more costly than the calculation of the endmember matrix. The results confirm that clustering the images with appropriate clustering methods and cluster numbers before endmember estimation and unmixing not only requires less calculation time than using the unclustered image, but additionally improves the results of both endmember and abundance estimation in the presence of noise.
更多查看译文
关键词
Hyperspectral image, unmixing, clustering, regularization, spatial information
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络