Comparison of linear and non-linear feature extraction on vegetation and oil spill hyperspectral images
Abstract
Description
A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Computer Science from Texas A&M University-Corpus Christi in Corpus Christi, Texas.
A hyperspectral image provides a multidimensional figure rich in data consisting of hundreds of spectral dimensions. For this research, the method of analysis for a hyperspectral image will consist of two different feature extraction algorithms: principal component analysis locally linear embedding. Analyzing the spectral and spatial information of such image with linear and non-linear algorithms will result in high computational time. In order to overcome this problem, this research proposes a system using a MapReduce-Graphics Processing Unit (GPU) model that can help analyze a hyperspectral image through the usage of parallel hardware and a parallel programming model, which will be simpler to handle compared to other low level parallel programming models. Additionally, Hadoop will be used as an open-source version of the MapReduce parallel programming model. The ultimate goal of this research is to provide a foundation for a simple and powerful system that is scalable and easily extend-able.
Computing Sciences
College of Science and Engineering
A hyperspectral image provides a multidimensional figure rich in data consisting of hundreds of spectral dimensions. For this research, the method of analysis for a hyperspectral image will consist of two different feature extraction algorithms: principal component analysis locally linear embedding. Analyzing the spectral and spatial information of such image with linear and non-linear algorithms will result in high computational time. In order to overcome this problem, this research proposes a system using a MapReduce-Graphics Processing Unit (GPU) model that can help analyze a hyperspectral image through the usage of parallel hardware and a parallel programming model, which will be simpler to handle compared to other low level parallel programming models. Additionally, Hadoop will be used as an open-source version of the MapReduce parallel programming model. The ultimate goal of this research is to provide a foundation for a simple and powerful system that is scalable and easily extend-able.
Computing Sciences
College of Science and Engineering