Hyperspectral sub-pixel target detection using hybrid algorithms and Physics Based Modeling
Note: imported from RIT’s Digital Media Library running on DSpace to RIT Scholar Works. Physical copy available through RIT's The Wallace Library at: G70.4 .I48 2005
This thesis develops a new hybrid target detection algorithm called the Physics Based-Structured InFeasibility Target-detector (PB-SIFT) which incorporates Physics Based Modeling (PBM) along with a new Structured Infeasibility Projector (SIP) metric. Traditional matched ?lters are susceptible to leakage or false alarms due to bright or saturated pixels that appear target-like to hyperspectral detection algo- rithms but are not truly target. This detector mitigates against such false alarms. More often than not, detection algorithms are applied to atmospherically com- pensated hyperspectral imagery. Rather than compensate the imagery, we take the opposite approach by using a physics based model to generate permutations of what the target might look like as seen by the sensor in radiance space. The development and status of such a method is presented as applied to the generation of target spaces. The generated target spaces are designed to fully encompass image tar- get pixels while using a limited number of input model parameters. Evaluation of such target spaces shows that they can reproduce a HYDICE image target pixel spectrum to less than 1% RMS error (equivalent re?ectance) in the visible and less than 6% in the near IR. Background spaces are modeled using a linear subspace (structured) approach characterized by basis vectors found by using the maximum distance method (MaxD). The SIP is developed along with a Physics Based Orthogonal Projection Op- erator (PBosp) which produces a 2 dimensional decision space. Results from the HYDICE FR I data set show that the physics based approach, along with the PB- SIFT algorithm, can out perform the Spectral Angle Mapper (SAM) and Spectral Matched Filter (SMF) on both exposed and fully concealed man-made targets found in hyperspectral imagery. Furthermore, the PB-SIFT algorithm performs as good (if not better) than the Mixture Tuned Matched Filter (MTMF).