SPECTROSCOPIC CALIBRATION AND QUANTITATION USING ARTIFICIAL NEURAL NETWORKS

This article demonstrates the application of artificial neural networks for nonlinear multivariate calibration using spectroscopic data. Neural networks consisting of three layers of nodes were trained by using the back-propagation learning rule. Sigmold output functions were used in the hidden layer to facilitate nonlinear fitting. Adjustable network parameters were optimized by using simulated data. The effect of random error In the concentration variables and In the response variables was investigated. The technique was tested by using real data, Including an example showing the determination of protein in wheat using near-infrared spectroscopic data and two examples showing the quantitation of the ingredients in pharmaceutical products using ultraviolet-visible spectroscopic data. © 1990, American Chemical Society. All rights reserved.