Classification performance of reduced color band hyperspectral imagery from JPEG 2000 decompressed data
Abstract
Hyperspectral imagery brings to remote sensing a whole new set of capabilities. Common images are represented in a computer by the compound of three color bands (red, green and blue). Hyperspectral image (HSI) can be viewed as an extension of a common image, beyond the limitations of human color perception. By being able to sense a greater range of bandwidth, we can distinguish the different elements present in the scene that could not be previously detected. Due to the characteristics of the HSIs, the analysis and processing of these types of files, generates high computational and memory costs. In order to reduce these costs, this thesis proposes to losslessly compress the HSI using the JPEG 2000 standard; and only decompress a small number of color bands to reconstruct the HSI. After the compress/decompress process, a classification algorithm is applied to the resulting HSI to measure the impact on the accuracy. One of the objectives of this thesis is to find the minimum number of color bands that achieves a hit rate greater than ninety percent in the classification, of aluminum-rich (Al-rich) and aluminum-poor (Al-poor) mica. Another objective is to establish a successive approach performed by two classifications; the first will establish the possible mica areas and the second will determine the type of mica (Al-rich or Al-poor). These two objectives have the goal to reduce the computational cost of the decompress/classify process, by reducing the amount of data computed.
Subject Area
Computer science|Remote sensing
Recommended Citation
Garces, Maximiliano, "Classification performance of reduced color band hyperspectral imagery from JPEG 2000 decompressed data" (2003). ETD Collection for University of Texas, El Paso. AAIEP10546.
https://scholarworks.utep.edu/dissertations/AAIEP10546