Remote Hyperspectral Imaging (HSI) is a powerful tool for identifying different materials on the surface of a large geographic area. Detection algorithms have been developed to detect specific target materials of interest in imaged scenes. Hyperspectral detection performance depends on the variability of the particular materials in a scene and the detection algorithm being used. Performance prediction of detection algorithms is important for understanding the capabilities and limitations of hyperspectral target detection. In this thesis, we present a robust and efficient technique for performance prediction of hyperspectral target detection algorithms.
Traditional methods for hyperspectral target detection performance prediction involve obtaining an analytic expression of the detector output’s statistical distribution or simulating data from an input statistical model. Except for simple cases, analytic expression of detector output are generally unavailable. Standard Monte Carlo (MC) simulation techniques are typically inefficient for estimating detector performance to the degree required in HSI applications. In this thesis, we use a modified MC technique known as Importance Sampling (IS) to design an efficient and robust simulation with the goal of predicting hyperspectral target detection performance.
The IS technique developed in this thesis is then used to predict the performance of actual hyperspectral data. It is shown that our IS technique is efficient and robust even in complicated data models.
- Professor Vinay Ingle (Advisor)
- Dr. Dimitris Manolakis (Advisor)
- Professor Purnima Makris