Advanced search
Start date

Application of Near Infrared spectroscopy and hyperspectral imaging to quantify the oil content and classify Brassica seeds

Full text
Maria Lucimar da Silva Medeiros
Total Authors: 1
Document type: Master's Dissertation
Institution: Universidade Estadual de Campinas (UNICAMP). Faculdade de Engenharia de Alimentos
Defense date:
Examining board members:
Renato Grimaldi; Raul Benito Siche Jara
Advisor: Douglas Fernandes Barbin

Brassica seeds are one of the most important sources of vegetable oils in the world and have numerous applications in the food, pharmaceutical and chemical industries. The oil content, the main quantitative parameter to assess its suitability for the market, is usually measured by traditional analytical techniques, which are time consuming, costly and destroy the sample. Vibrational techniques, such as near-infrared spectroscopy (NIR) and hyperspectral imaging (NIR-HSI), allow simple, fast, and accurate quantification of chemical components in food, without destroying the sample and without generating chemical residues and can be applied as an alternative to conventional methods. Therefore, the present study aimed to investigate the potential of spectra acquired in a portable equipment (NIRS) (900 - 1700 nm) and in a hyperspectral camera (NIR-HSI) (928 - 2524 nm), together with chemometrics, species authentication and prediction of oil content in intact Brassicas seeds. Principal Component Analysis (PCA) was used as an exploratory analysis of the data. Discriminant Analysis by Partial Least Squares (PLS-DA) was applied to discriminate between Brassica napus, Brassica rapa and Brassica juncea species and proved to be efficient, with accuracy between 75.0 - 93.6% and 91.0 - 100% for NIRS and NIR-HSI data. Partial Least Squares regression (PLS) was used to predict the oil content in intact seeds and demonstrated good predictive ability, especially for the pre-processed NIR spectra with first derivative and for the raw NIR-HSI spectra, where they were obtained determination coefficients of 0.682 and 0.764, mean square errors in the prediction less than 1.0 and RPD of 2.0 and 2.2, respectively. The selection of relevant wavelengths from the PLS interval algorithm (iPLS) increased the discriminative capacity of the NIRS model based on the spectra smoothed with Savitzky-Golay, where an accuracy of 94.9% was achieved, and provided an improvement in the results of the PLS calibration for the NIR-HSI spectra, especially when pre-processed with second derivative, reaching RPD of 2.31. The results obtained in this study showed that both NIR spectroscopy devices, combined with chemometric tools, can be applied to discriminate seeds of the analyzed Brassicas species and to provide a quick estimate of the oil content in the seed (AU)