Detection and classification of subpixel spectral signatures in hyperspectral image sequences

Hyperspectral imaging spectrometer data is made up of hundreds of spatially registered images taken contiguously over a large wavelength region with high ({dollar}<{dollar}10 nm) spectral resolution. Each pixel in a hyperspectral image sequence is an observation vector which represents the reflected energy spectrum of the materials within the spatial area covered by the pixel. The combination of high spectral resolution and continuous wavelength coverage provides the opportunity to detect and classify the surface materials contained in each pixel based on their spectral signatures.; The requirement to detect and classify surface materials is a common theme in a variety of earth remote sensing applications. An additional requirement for most applications is to reduce the data volume/dimensionality, without loss of critical information, so that it can be processed efficiently and assimilated by a human analyst. This dissertation describes the development of three new techniques which address these requirements.; The first technique is the orthogonal subspace projection (OSP) operator. This operator is shown to be an optimal interference rejection process in the least squares sense, as well as a maximum signal-to-noise ratio operator for detection of a spectral signature of interest which occurs in a subpixel mixture with multiple interfering signatures not of interest. The subpixel mixture problem occurs when multiple materials with unique spectral signatures are present within the spatial coverage of a single pixel.; The second technique is an extension of the OSP operator which is focused on the problem of detecting low probability subpixel target signatures when the prevailing background signatures are unknown. This technique is especially useful for detection of man-made objects in naturally occurring backgrounds, and can be applied to the detection of sparse vegetation, scarce mineral deposits and other poorly exposed geologic features.; Finally, a general approach for detection of spectral signatures in unknown backgrounds is developed. The constrained energy minimization (CEM) operator minimizes the total energy in a hyperspectral image sequence while the output of the operator is constrained to have a unity response when applied to the signature of interest.