NIR spectrometer used for material modeling with neural networks

Abstract

Near infrared multi-spectral image analysis is a tool used for non-destructive determination of biological material properties. In this investigation a custom built imaging spectrometer is constructed and used for the image spectra instrumentation and tests are performed on this instrument to determine its spectral resolution and spectral range; a biological data set (moisture in potato crisps) is then captured using this instrument and this data set is modeled using near infrared multi-spectral image analysis. A common problem with near infrared multi-spectral quantitative image measurements is light scatter and light non-linearity resulting from sample shape contours/curvatures and optical aberrations from optical component selection/layout. In this paper we detail an imaging spectrometer and the use of orthogonal signal correction pre- processing combined with a neural network full spectrum model for measurement of material property