OPTIMAL ASSOCIATIVE MEMORY FOR BACKGROUND CORRECTION OF SPECTRA

A novel artificial neural network has been devised and is evaluated fo r the background correction of single-scan infrared (IR) spectre. An o ptimal associative memory (OAM) is an enhanced bidirectional associati ve memory (BAM). Factoring the weight matrix allows OAMs to be used wi th high-resolution data on a desktop computer. IR spectroscopy provide s a rigorous and practical challenge for evaluating background correct ion. IR single-scan background spectra are stored in the associative m emory. Single-scan sample spectra are used to retrieve the best fittin g background scans. The OAM uses an internal Gram-Schmidt calculation and does not require orthogonal data. The associative properties of th e OAM allow background scans not stored in the memory to be modeled. T he memories were evaluated with 2-cm(-1) resolution IR spectra. Quanti tative analyses of 2-octanone/toluene solutions were used to evaluate the OAM with regard to accuracy and linearity. In both cases of univar iate and multivariate calibrations, the OAM-corrected spectra furnishe d better calibration models than those obtained from conventional IR a bsorbance spectra.