Meta-scalable discriminate analytics for Big hyperspectral data and applications

Recent technology developments in hyperspectral sensing has made it possible to acquire several hundred spectral bands that cover the electromagnetic spectrum of an observational scene in a single acquisition. The resulting hyperspectral data cube contains a large volume of spatial-spectral information. It has several concrete and special characteristics such as being multi-source, multi-scale, high dimensional and nonlinear. The hyperspectral video with temporal information further increases the data generation velocity and volume which lead to the Big data challenges especially in remote sensing applications. We term this type of Big data as Big hyperspectral data to differentiate it from the Big data generated from internet and multimedia-based sources. This paper presents a novel data computation framework for Big hyperspectral data discriminate analytics. This framework consists of some essential modules like tree-based divide-conquer (Tree-DC) mechanism, hierarchical spatial-spectral domain (HSSD) decomposition, global scalable and locally fast discriminative analytics (GSLF-DA), tree-based divide-conquer-merge (DCM), and temporal hyperspectral data decomposition. The challenge of the framework is to sustain the divide-conquer scalability for implementation on rapidly evolving parallel computing architectures i.e., transforming the divide-conquer mechanism to be meta-scalable. Moreover, the discriminate analytics in conjunction with the proposed mechanism can give the optimal solution in the final merging stage. Experiments are performed to validate the performance of the mechanisms in the framework.