Predicción de contaminantes físicos en conserva de bofe de res utilizando imágenes hiperespectrales
No Thumbnail Available
Date
2022
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad Nacional de Trujillo
Abstract
La producción de conservas de bofe de res viene presentando volúmenes de consumo importantes en los ultimos años, debido a la demanda de los programas sociales de lucha contra la anemia y desnutrición infantil, que cumplen con parámetros de calidad e inocuidad. Estos parámetros pueden verse comprometidos por la presencia de contaminantes como restos de latex de guante, plástico y cabello, que están presentes a lo largo del proceso y que se pueden hallar en el producto final. Por lo tanto, el objetivo de este trabajo fué aplicar el uso de imágenes hiperespectrales y analisis multivariante para discriminar y cuantificar muestras puras de conserva de bofe de res de muestras contaminadas con restos de guante, plástico y cabello, a diferentes concentraciones (0, 0.29, 0.57, 0.85, 1.14, 1.42, 1.71, 2.00, 2.28%) y con 10 repeticiones por concentración, para cada contaminante. El acondicionamiento de las muestras se realizó similar al trabajo en planta de procesos. Las imágenes hiperespectrales se adquirieron de forma individual en el rango espectral de infrarrojo cercano de 900 – 1710 nm. Se selección la región de interés (ROI) y se obtuvo los espectros medios, los cuales fueron usados para realizar el análisis multivariado. Se realizó un preprocesamiento espectral con técnicas de corrección utilizando la combinación de centrado en la media (MC) y variable normal estándar (SNV) y la primera derivada de Savitzky-Golay, para reducir el error por dispersión de la luz y ruido aleatorio. Se identifico las bandas espectrales que representan a los principales compuestos de bofe puro y los compuestos de sus principales contaminantes El análisis estadístico multivariado consistió en un análisis cualitativo de componentes principales (PCA) donde se logró diferenciar entre las clases de conserva de bofe puro y conserva de bofe contaminado, con los dos primeros componentes principales. Para el método de cuantificación de regresión de mínimos cuadrados parciales (PLS-R), se seleccionó 70 % de muestras de calibración y 30 % de validación. Los modelos PLS-R presentaron una variada capacidad predictiva, siendo el mejor modelo la contaminación con restos de guantes (R_P^2 > 0.90 y RPD > 3.20) y el modelo más bajo fue el de contaminación de mezcla de contaminantes 〖(R〗_P^2 > 0.68 y RPD > 1.80). En general el uso NIR-HSI es efectivo para detectar los principales contaminantes de conserva de bofe de res, siendo el mejor identificado la contaminación con plástico.
ABSTRACT The production of canned beef bofe has been presenting significant consumption volumes in recent years, due to social programs to combat anemia and child malnutrition, which needs to meet the minimum parameters of quality and safety. These parameters can be compromised by the presence of contaminants such as latex glove, plastic, and hair residues, which are made up of dangerous substances. Therefore, the objective of this work was to apply the use of hyperspectral imagery and multivariate analysis to discriminate and quantify pure samples of beef meat preserve from samples contaminated with glove, plastic, and hair residues, at different concentrations (0, 0.29, 0.57, 0.85, 1.14, 1.42, 1.71, 2.00, 2.28%) and with 10 repetitions per concentration, for each contaminant. The conditioning of the samples was carried out like the work in the plant. Hyperspectral images were acquired individually in the near infrared spectral range of 900-1710 nm. The region of interest (ROI) was selected, and the mean spectra were obtained, which were used to perform the multivariate analysis. Spectral preprocessing was performed with correction techniques using the combination of centered on the mean (MC) and standard normal variable (SNV) and the first derivative of Savitzky-Golay, to reduce the error due to light scattering and random noise. The spectral bands that represent the main compounds of pure bofe and the compounds of its main pollutants were identified. of contaminated bofe, with the first two main components. For the partial least square’s regression (PLS-R) quantification method, 70% of samples were selected for calibration and 30% for validation. The PLS-R models presented a varied predictive capacity, the best model being contamination with remains of gloves (R_P^2 > 0.90 y RPD > 3.20) and the lowest model was the contamination of a mixture of contaminants 〖(R〗_P^2 > 0.68 y RPD > 1.80). In general, the use of NIR-HSI is effective in detecting the main contamination of beef canned beef, resulting as the best identified contamination with plastic.
ABSTRACT The production of canned beef bofe has been presenting significant consumption volumes in recent years, due to social programs to combat anemia and child malnutrition, which needs to meet the minimum parameters of quality and safety. These parameters can be compromised by the presence of contaminants such as latex glove, plastic, and hair residues, which are made up of dangerous substances. Therefore, the objective of this work was to apply the use of hyperspectral imagery and multivariate analysis to discriminate and quantify pure samples of beef meat preserve from samples contaminated with glove, plastic, and hair residues, at different concentrations (0, 0.29, 0.57, 0.85, 1.14, 1.42, 1.71, 2.00, 2.28%) and with 10 repetitions per concentration, for each contaminant. The conditioning of the samples was carried out like the work in the plant. Hyperspectral images were acquired individually in the near infrared spectral range of 900-1710 nm. The region of interest (ROI) was selected, and the mean spectra were obtained, which were used to perform the multivariate analysis. Spectral preprocessing was performed with correction techniques using the combination of centered on the mean (MC) and standard normal variable (SNV) and the first derivative of Savitzky-Golay, to reduce the error due to light scattering and random noise. The spectral bands that represent the main compounds of pure bofe and the compounds of its main pollutants were identified. of contaminated bofe, with the first two main components. For the partial least square’s regression (PLS-R) quantification method, 70% of samples were selected for calibration and 30% for validation. The PLS-R models presented a varied predictive capacity, the best model being contamination with remains of gloves (R_P^2 > 0.90 y RPD > 3.20) and the lowest model was the contamination of a mixture of contaminants 〖(R〗_P^2 > 0.68 y RPD > 1.80). In general, the use of NIR-HSI is effective in detecting the main contamination of beef canned beef, resulting as the best identified contamination with plastic.
Description
Keywords
Banda espectral, Contaminantes físicos, Análisis de imágenes y datos