Efficient sparse signal recovery of remote sensing data: a classification method for hyperspectral image data

Kurzfassung

Nowadays the concern of finding an efficient algorithm that can answer some of the open questions in big data analysis and mining has been gradually arose. Such questions can be regarded by the question of representing the data in a meaningful way in which the most useful information highlighted. Therefore, the motivation of answering these questions encourage this thesis to develop a principle classification algorithm called Efficient sparse signal recovery for big data representation for a classification task. In this thesis, we develop a classification principle algorithm that is based on the sparse coding for the classification of given test pixel from a hyperspectral image. Hyperspectral imagery in remote sensing domain has the characteristic of big data in terms of velocity, verity and volume. This data is a set of non-homogenous system that expose the ill-posed problem. Thus, a robust and efficient algorithm must be developed to treat such data effectively. Sparse representation draws a great attention in hyperspectral image representation and analysis. Employing sparsity-based model involved two main problems. Firstly, the problem of the representation of an informative dictionary, and secondly the issue of implementing a proper optimization problem that can effectively solve the objective function. This thesis focuses on the latter aspect while the dictionary issue is also tackled by proposing a Geometric dictionary. There have been many algorithms for finding the optimized minimum of the well-known objective functionals “least square ” with