Imagens hiperespectrais para o controle da qualidade de alimentos: híbridos de graos de cacau e vida de prateleira de sementes de chia

Author(s)

Luis Jam Pier Cruz Tirado

• Document type: Master's Dissertation
• Press: Campinas, SP.
• Institution: Universidade Estadual de Campinas (UNICAMP). Faculdade de Engenharia de Alimentos
• Defense date: 2020-09-25

Examining board members

Douglas Fernandes Barbin; María Nuria Aleixos Borrás; Jose Blasco Ivars

• Advisor: Douglas Fernandes Barbin

Abstract

Hyperspectral imaging (HSI) enables simultaneous acquisition of spectral and spatial information. In this work, HSI was used for quality control of agricultural products, which includes the authentication of cocoa bean hybrids and the estimation of shelf-life of chia seeds. Regarding the chia seeds study, samples were stored at 25, 35 and 45 ° C for 180 days, for accelerated shelf life analyzes. From time to time, chia samples were removed from storage to acquire hyperspectral images (900 - 2500 nm), acidity analysis, and fatty acid profile. The objective was to use hyperspectral images and multivariate analysis to develop a methodology for estimating the shelf-life of chia seeds, called Multivariate Accelerated Shelf Life Testing (MASLT). Principal Component Analysis (PCA) was used to study the variability during storage, and then, the PC scores were used to model the kinetics and estimate the parameters of the Arrhenius Equation, and finally to estimate the shelf life. Furthermore, for the first time a new strategy was proposed to validate this methodology, which we called "Re-sampling", where the samples from the validation set were projected onto the calibration set with a reasonable number of iterations. PC1 scores and kinetic charts were built fitting the time-related PC1 scores versus time by a fused kinetic model (R2 > 0.85). The spectra of chia seeds where acidity increased at 75% from initial value were used to calculate the cut-off value (-0.9853). The shelf life estimations were 1300, 798 and 90 days for chia seeds stored at 25, 35 and 45 °C, respectively. For the first time, a reliable methodology is proposed to validate that all samples were correctly predicted using PC1 scores. In the second study, cocoa beans hybrids (five) were grown and processed under the same conditions in CEPLAC (Medicilândia, Para, Brazil). The cocoa beans were then transported to the Wallonie Research Center (Belgium), where hyperspectral images in the 1100 - 2500 nm range were acquired. Partial least square discriminant analysis (PLS-DA) and Support vector machine (SVM) was implemented to classify cocoa bean hybrids, (1) two classes of hybrids and (2) five classes of hybrids. Additionally, a new set of images was used for external pixel-to-pixel validation. The results showed that PLS-DA and SVM demonstrate comparable results for two-class (hybrids) models, but SVM (3.8–23.1% prediction error) was superior to PLS-DA (4.4–34.4% prediction error) when all five classes (hybrids) were included in a model. Pixel-to-pixel prediction results on a set of external images showed a correct classification rate of 50 - 100%. The results for both the two-class models and the five-class model were comparable with polymerase chain reaction techniques. The results show the potential of HSI for quality control of agricultural products, both for authentication and estimation of shelf life (AU)

• FAPESP's process: 18/02500-4 - Food analyses using NIR spectral imaging
• Grantee: Luis Jam Pier Cruz Tirado
• Support Opportunities: Scholarships in Brazil - Master