Automated Infrared Detection of Organophosphorus Compounds in MulticomponentSolutions

摘要

In-situ monitoring of the environment in peacetime and wartime is required oftoday's technology with particular concern toward organophosphorus pesticides and chemical warfare agents. Of the many tools available for monitoring, infrared spectrometry combined with fiber optic sensors is one method of remote sensing which shows promise. This thesis models the problem of automatically detecting organophosphorus chemicals in water and automatically specifying what chemical is present. Computer generated infrared spectra are modeled as well as sensor characteristics and Computer generated infrared spectra are modeled as well as sensor characteristics and instrument noise over various binary mixture concentrations. A standard pattern recognition data reduction technique known as Principle Components Analysis is employed to compute meaningful features for detection of target chemicals in water. Methods of detection of the target chemical include the k-nearest neighbor classifier and Parzen window/maximum a-posteriori probability decision rule. By systematically varying chemical amounts, the order of magnitude of the limit of detection is determined. To determine the detection error bound beyond a simple point estimate of the error, Bayes error bounding is implemented. Once the lower range of detection is determined, k-nearest neighbor tests are done to determine classification of specific target chemicals. jg p13.