METHODS OF ANALYSIS AND CLASSIFICATION OF THE COMPONENTS OF GRAIN MIXTURES BASED ON MEASURING THE REFLECTION AND TRANSMISSION SPECTRA

• Main Authors: Artem O. Donskikh*, Dmitry A. Minakov, Alexander A. Sirota, Vladimir A. Shulgin
• Format: Article
• Language: English
• Published: Alma Mater Publishing House "Vasile Alecsandri" University of Bacau 2017-10-01
• Series: Scientific Study & Research: Chemistry & Chemical Engineering, Biotechnology, Food Industry
• Subjects: grain mixtures; neural networks; pattern recognition; seed quality; spectral analysis
• Online Access: http://pubs.ub.ro/?pg=revues&rev=cscc6&num=201703&vol=3&aid=4618

The paper considers methods of classification of grain mixture components based on spectral analysis in visible and near-infrared wavelength ranges using various measurement approaches - reflection, transmission and combined spectrum methods. It also describes the experimental measuring units used and suggests the prototype of a multispectral grain mixture analyzer. The results of the spectral measurement were processed using neural network based classification algorithms. The probabilities of incorrect recognition for various numbers of spectral parts and combinations of spectral methods were estimated. The paper demonstrates that combined usage of two spectral analysis methods leads to higher classification accuracy and allows for reducing the number of the analyzed spectral parts. A detailed description of the proposed measurement device for high-performance real-time multispectral analysis of the components of grain mixtures is given.