Pattern Recognition Methods for FTIR Remote Sensing.
NIOSH 1991 Feb:549-558
Pattern recognition techniques were described that allow the implementation of an effective decision making algorithm for use in analyzing filtered interferogram segments. The usefulness of this method was demonstrated through a large quantity of passive Fourier transform infrared (FTIR) remote sensing data. The results confirmed that a short interferogram segment can be used for the reliable detection of target analytes from passive FTIR data. The combination of digital filtering and pattern recognition techniques allowed this detection algorithm to be implemented. This achievement made possible the design of a new generation of passive FTIR sensors based on the short scan interferometer concept. Two new general purpose algorithms were presented for use in pattern recognition analyses. The training set selection algorithm described can be used to select training sets for use with any pattern recognition method. The method outperformed pattern selection strategies based on random sampling of a pool of candidate patterns. The techniques would be applicable to any pattern recognition problem in which the interface between the data classes is complex.
Spectrographic-analysis; Analytical-methods; Analytical-chemistry; Chemical-analysis; Analytical-processes; Mathematical-models; Air-quality-monitoring;
Field Screening Methods for Hazardous Wastes and Toxic Chemicals. Second International Symposium, February 12-14, 1991. Sponsored by U.S. EPA; U.S. DOE; U.S. Army Toxic and Hazardous Materials Agency; U.S. Army Chemical Research Development and Engineering Center; U.S.A.F.; Florida State Univ.; National Environmental Technology Applications Corp.; and NIOSH