Hyperspectral imagery (HSI) remote sensing is the simultaneous acquisition of hundreds of coregistered images of a scene by sampling many contiguous, narrow-band regions of the electromagnetic (EM) spectrum. An HSI pixel is thus a spectrum: an n-dimensional (nD) vector where n is equal to the number of bands or samples of the EM spectrum. The spectrum may also be represented as a point in an nD hyperspace. Common HSI sensors available today have on the order of 128 to 224 bands. HSI data sets are commonly referred to as cubes: data structures comprised of two spatial dimensions (image plane) and one spectral dimension (wavelength sampling of the EM spectrum). HSI cubes contain scientific data representing the interaction of photons of light with matter. They are large both in terms of the number of bytes recorded and in terms of the dimensionality of hyperspace occupied by the multivariable nature of the spectral information. Efficient information extraction, data/product transmission, and data/product storage may benefit from preprocessing techniques that attempt to reduce HSI dimensionality. An introduction to HSI is presented followed by discussions on defining HSI dimensionality, techniques to measure it, and techniques to reduce it. The utility of dimensionality reduction is presented within the framework of existing techniques for information extraction from HSI.
Presentation (PowerPoint File)
Video of Talk (RealPlayer File)
Copyright. All Rights Reserved.