Examination of stochastic and ordered methods to select optical filters for discrimination between chemical vibrational absorption bands

摘要

Recent developments have shown that an optical filter-based sensing approach, inspired by human color vision, is capableof high-confidence discrimination between chemicals with similar infrared vibrational absorption bands. A key designpoint for this technique lies in the selection of the optical filters, which provide good discrimination between chemicals.Filter selection is also intrinsically tied to the classification method employed for the discrimination itself. Thus, it isimperative that the classification method or methods to be used are well understood and that mathematical means exist tocompare the discrimination results provided by independent sets of optical filters. To meet this challenge, we are examiningmeans to assign cost values to each set of optical filters for a given associated classification method. In this effort, the costvalue used is the volume formed by three unique discrimination vectors. This method is developed from machine learningapproaches, which define cost functions for stochastic optimization routines. We discuss multiple computational methodsto discriminate between chemicals with similar infrared vibrational absorption bands using unique infrared (IR) tristimulusvalues for each chemical. These IR-tri-stimulus values are determined by the interaction with the chemicalabsorption bands and three individual optical IR band-pass filters. Methods to determine the associated cost for variousselections of these IR band-pass filters and associated mathematical operations are described and compared for thecomputational methods. We discuss the methods employed to select the IR optical filters and discuss how the flexibilityof this approach demonstrates the power of this biomimetic sensing method.