A Proof of Concept Study for the Parameters of Corn Grains Using Digital Images and a Multivariate Regression Model
- PDF / 898,928 Bytes
- 5 Pages / 595.276 x 790.866 pts Page_size
- 7 Downloads / 204 Views
A Proof of Concept Study for the Parameters of Corn Grains Using Digital Images and a Multivariate Regression Model Vanessa Rodrigues de Camargo 1 & Lucas Janoni dos Santos 1 & Fabíola Manhas Verbi Pereira 1
Received: 7 June 2017 / Accepted: 20 August 2017 # Springer Science+Business Media, LLC 2017
Abstract In this method, a numerical matrix comprised of ten color scales (RGB, HSV, L, and rgb) as independent variables from digitalized images was used as a proof of concept for the prediction of the mass, apparent volume, and bulk density parameters of grains for quality control considering postharvest purposes. The goal was to develop a high throughput multivariate regression model using partial least squares (PLS) combined with the information from color images to assess the raw product. The data set of external samples was successfully evaluated with standard error of cross-validation (SECV) values of 1.23 g (16.4–28.9), 2.03 cm3 (20.5–40.5), and 0.018 g cm−3 (0.68–0.85) for the mass, apparent volume, and bulk density, respectively. Keywords Corn grains . Digital images . Direct analysis . Quality control . Chemometrics
Introduction Image analysis converted into mathematical arrays can be performed with the aid of several chemometric tools (Russ, 1992; Wojnar, 1999; Liu & MacGregor, 2007; Pereira & Bueno, 2007). In this study, the principle will be an emphasis on the Electronic supplementary material The online version of this article (https://doi.org/10.1007/s12161-017-1028-6) contains supplementary material, which is available to authorized users. * Fabíola Manhas Verbi Pereira [email protected] 1
Instituto de Química de Araraquara, Universidade Estadual Paulista (UNESP), Rua Professor Francisco Degni, 55, Araraquara, SP 14800-060, Brazil
partial least squares (PLS) regression method (Geladi et al., 2004). Walker and Panozzo (Walker & Panozzo, 2012) evaluated three different mathematical models for measuring the volume and bulk density of a barley grain using an ellipsoid approximation from a two-dimensional digital image. They achieved correlations to grain features of 0.97 and 0.63 for the volume and bulk density, respectively. However, it is not reported how the PLS is able to perform for the prediction of these properties using a single model. By means of the red, green, and blue (RGB) scale and chemometrics, Borin et al. (Borin et al., 2007) accomplished the quantification of lactobacilli colonies in commercial fermented milk using digital images provided by a household scanner. The authors applied one-dimensional vector data based on the frequency distribution of values of curves (histograms) for the three colors (RGB). Two models have been developed: (i) the first was nonlinear using LS-SVM (support vector machine least squares) and (ii) the second was linear based on PLS. Both models were used to calculate the number of lactobacilli colonies. The relative error for the quantification was approximately 10%. This fact indicated that the proposed method can be used for automated counting of this typ
Data Loading...
