Hybrid pattern recognition method using evolutionary computing techniques applied to the exploitation of hyperspectral imagery and medical spectral data

摘要

Abstract: Hyperspectral image sets are three dimensional datavolumes that are difficult to exploit by manual meansbecause they are comprised of multiple bands of imagedata that are not easily visualized or assessed. GTEGovernment Systems Corporation has developed a systemthat utilizes Evolutionary Computing techniques toautomatically identify materials in terrainhyperspectral imagery. The system employs sophisticatedsignature preprocessing and a unique combination ofnon- parametric search algorithms guided by a modelbased cost function to achieve rapid convergence andpattern recognition. The system is scaleable and iscapable of discriminating and identifying pertinentmaterials that comprise a specific object of interestin the terrain and estimating the percentage ofmaterials present within a pixel of interest (spectralunmixing). The method has been applied and evaluatedagainst real hyperspectral imagery data from the AVIRISsensor. In addition, the process has been applied toremotely sensed infrared spectra collected at themicroscopic level to assess the amounts of DNA, RNA andprotein present in human tissue samples as an aid tothe early detection of cancer. !24