Analysis and classification of peanuts with fungal diseases based on real-time spectral processing

• Published in: Food additives & contaminants. Part A, Chemistry, analysis, control, exposure & risk assessment Chemistry, analysis, control, exposure & risk assessment, 2022-05, Vol.39 (5), p.990-1000
• Main Authors: Lavrinenko, Igor A., Donskikh, Artem O., Minakov, Dmitriy A., Sirota, Alexander A.
• Format: Article
• Language: English
• Subjects: Aflatoxins - analysis; Algorithms; Arachis - chemistry; Aspergillus flavus; Classification; Contamination; Fluorescence; Food Contamination - analysis; Fungal diseases; fungal diseases of peanut; image classification; Kernels; Light sources; luminescence spectroscopy; Luminous intensity; Machine learning; Mycoses; Neural networks; Nuts; on-stream express analysis; Peanuts; Performance evaluation; Real time; Red shift; Spectra; Spectroscopy; Spectrum analysis; Ultraviolet radiation; Violet laser; Wavelength
• ISSN: 1944-0049; 1944-0057
• DOI: 10.1080/19440049.2021.2017001

The study presents an approach to the analysis and classification of peanuts performed in order to detect kernels with fungi diseases, i.e. kernels prone to contamination with mycotoxigenic Aspergillus flavus (Aspergillus parasiticus). The aim of this study was to evaluate the effectiveness of luminescent spectroscopy with a violet laser (405 nm wavelength) as the excitation source of the fluorescence when applied for real-time detection of mould in peanuts performed by means of multispectral processing based on machine learning methods. We suggest a laboratory unit used to form, register, and process the luminescence spectra of peanuts in visible and near-infrared wavelength ranges in the real-time mode. The study demonstrated that contaminated peanuts have increased luminous intensity and show a redshift in the fluorescence peaks of the contaminated samples as compared to the pure ones. The difference in the fluorescence spectra of pure and contaminated kernels is compatible with the results obtained when traditional UV-light sources are used (365 nm). To classify peanuts by their spectral characteristics, neural network algorithms were used combined with dimensionality reduction methods. The paper presents the probabilities of incorrect recognition of the peanuts' type depending on the number of relevant secondary features determined when reducing the dimensionality of the initial data. When 10 spectral components were used, the error ratios were 0.7% or 0.3% depending on the method of reducing the dimensionality of the initial data.