Distributed Processing of Hyperspectral Images

This paper examines several hyperspectral data processing algorithms designed for a distributed computing environment. Due to the large size, hyperspectral data requires long computational times to process. In a distributed environment, the processing can be split into several components, many of them being executed simultaneously, thus leading to increased time efficiency. The algorithms are derived from previously introduced feature extraction methods. The first is based on Principal Component Analysis (PCA). When applied to multidimensional data, PCA linearly transforms them such that the resulting components are uncorrelated and their variance maximized. In the context of hyperspectral data, PCA provides an efficient way for unsupervised feature extraction. The second, derived from Independent Component Analysis (ICA) is used to solve the linear unmixing problem. In ICA, given a linear mixture of statistical independent sources, the goal is to recover these components by producing an unmixing matrix. In multispectral/hyperspectral imagery, the separated components can be associated with features present in the image, the source separation algorithm projecting them in different image bands. PCA and ICA based methods have been employed for target detection and classification of hyperspectral images. However successful the methods, when applied to hyperspectral data, they yield relatively high execution times. Their time efficiency is improved by designing them for a distributed environment. The development of the distributed algorithms as well as issues related to distributed modeling of hyperspectral data are taken in consideration and will be presented. The effectiveness of the proposed algorithms was tested on AVIRIS and HYDICE data. Preliminary results indicate that, while the accuracy is preserved, the new algorithms provide considerable speed-up in processing.