Optimization Methods for Analyte Recognition from Optical Sensor Arrays

Abstract : This project focused on the analysis of an artificial olfactory system. Light at single wavelengths is piped down a doped fiber optic causing the dyes to fluoresce. As chemicals are drawn across the doped fiber optics, they adhere temporarily to the dye and change the color of the fluorescing dye. The response is transient as the chemicals dissipate. This creates a 19-dimensional time series for each chemical species. The time series is characteristic of the stimulating chemical species. The military application of such a system extends to detecting land mines and also detecting leaking hydrocarbons from damaged military systems such as tanks and helicopters. We have demonstrated that graphical methods allow for substantial discriminating capability. Current work involves development of what we have called a d-tour, an optimization technique for finding maximal discrimination capability among a number of chemical species. The idea is to create a distance metric based on L(p)-spaces with smoothers and weight functions. The d-tour is constructed in the same way as a grand tour, but in addition to touring on the data, we tour on the parameters of the smoother, of the weight function and on p, the exponent of the L(p)-space.