Dr. Wesley Snyder
of the ECE Department of NC State has been granted $110,000 for research on On-the-fly Scene-dependent Automatic Target Recognition (ATR) using hyperspectral and multispectral imaging data.
A hyperspectral imager produces an image with a hundred or more spectral measurements at each pixel. A multispectral imager on the other hand may sample the spectrum in only three to ten bands. Given hyperspectral data, one can synthesize multispectral data by simply integrating over the appropriate portion of the spectrum at each point.
One would think that a pattern classifier based on multispectral data would perform poorly as compared to the same type of classifier based on hyperspectral data, since surely information is being lost in going from many measurements to few. However, we know that the hyperspectral data is highly correlated and that much research has shown that the data has many fewer degrees of freedom than the number of individual bands.
This work will show how to use the hyperspectral data to it fullest advantage to detect and classify a target, then design a lower dimensional multispectral system that can perform the same task using fewer resources that can be implemented on a smaller platform.