Hyperspectral imagery analysis using neural network techniques

摘要

Every material has a unique electromagnetic reflectance/emission signature which can be used to identify it. Hyperspectral imagers, by collecting high spectral resolution data, provide the ability to identify these spectral signatures. Utilization and exploitation of hyperspectral data is challenging because of the enormous data volume produced by these imagers. Most current processing and analyzation techniques involve dimensionality reduction, during which some information is lost. This thesis demonstrates the ability of neural networks and the Kohonen Self-Organizing Map to classify hyperspectral data. The possibility of real time processing is addressed. (AN)