Hyperspectral imaging sensors are a relatively new type of camera that can capture several hundred channels of light over a wide range of wavelengths; by comparison, a standard color camera captures 3 channels (red, blue and green) over a relatively narrow wavelength range. The increased number of channels in hyperspectral data leads to a corresponding growth in the amount of information that can be extracted from an image. The tradeoff is an explosion in both the size (typical images can be on the order of 1 gigabyte) and complexity of the image data.
This talk presents a basic introduction to the mathematical models and algorithms that have been used to study hyperspectral image data. I begin with a brief introduction to the physics behind the data, and show how the physics imparts a great deal of mathematical structure to the data. I then present a survey of the various geometrical models (both linear and non-linear) that researchers are using to exploit this structure and better understand the information present in a given image.