Detección de carne de Alpaca (Vicugna pacos) adulterada usando espectrómetro portátil (NIRS) e imágenes hiperespectrales (HSI)
No Thumbnail Available
Date
2023
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad Nacional de Trujillo
Abstract
El objetivo de este trabajo fue el entrenar y validar modelos predictivos basados en espectrómetro NIR portátil e imágenes hiperespectrales NIR (NIR-HSI) con análisis quimiométrico para detectar adulteración en carne de alpaca. El estudio se realizó en carne de alpaca molida, adulterada con carnes de res, cerdo y pollo en un rango de 0% -50% (p/p). Se obtuvieron espectros usando ambos dispositivos NIR en la región espectral 900 – 1700 nm. Entonces, se utilizaron herramientas quimiométricas como análisis PCA, modelamiento SIMCA y PLSR para analizar los espectros. PCA capturó la varianza espectral, entonces fue posible agrupar los espectros de acuerdo con la clase (alpaca, alpaca + res, alpaca + cerdo, alpaca + pollo). Comparando ambos dispositivos, fue posible observar que NIR-HSI diferencia mejor las muestras de alpaca pura y alpaca adulterada comparado con el NIR portátil. Posteriormente, SIMCA fue utilizado para crear un modelo de clasificación para distinguir carne de alpaca pura (clase objetivo) y carne de alpaca adulterada. Los resultados de SIMCA fueron comparables para ambos dispositivos, con sensibilidad y especificidad > 95%. Los modelos PLSR se realizaron utilizando espectros completos y longitudes de onda seleccionadas por iPLS. Los mejores resultados se obtuvieron con las imágenes HSI, con coeficiente de predicción R2p (0.980, 0.917 y 0.981), relación rendimiento a desviación RPD (8.39, 3.82 y 7.96) y RMSEP de 1.87, 4.118 y 1.97, para muestras de alpaca adulterado con carne de res, cerdo y pollo, respectivamente. Además, los resultados obtenidos mediante el espectrómetro NIR portátil también son buenos resultados con coeficiente de predicción R2p (0.91, 0.82 y 0.83), relación rendimiento a desviación RPD (3.74, 2.64 y 2.73) y RMSEP de 4.10, 5.81 y 5.61. Finalmente, nuestros resultados muestran que ambos dispositivos NIR son una alternativa para la detección de carne de alpaca adulterada.
ABSTRACT This work aimed to train and validate predictive models based on a portable NIR spectrometer and NIR hyperspectral images (NIR-HSI) with chemometric analysis to detect adulteration in alpaca meat. The study was carried out on ground alpaca meat, adulterated with beef, pork and chicken in a range of 0% -50% (p/p). Spectra were obtained using both NIR devices in the spectral region 900 – 1700 nm. Then, chemometric tools such as PCA analysis, SIMCA modeling and PLSR were used to analyze the spectra. PCA captured the spectral variance, so grouping the spectra according to class (alpaca, alpaca + beef, alpaca + pig, alpaca + chicken) was possible. Comparing both devices, it was possible to observe that NIR-HSI better differentiates pure alpaca and adulterated alpaca samples compared to portable NIR. Subsequently, SIMCA was used to create a classification model to distinguish pure alpaca meat (target class) and adulterated alpaca meat. SIMCA results were comparable, with sensitivity and specificity > 95% for both devices. PLSR models were performed using full spectra and selected wavelengths by iPLS. The best results were obtained with the HSI images, with R2p prediction coefficient (0.980, 0.917 and 0.981), yield to RPD deviation ratio (8.39, 3.82 and 7.96) and RMSEP of 1.87, 4.118 and 1.97, for samples of alpaca adulterated with beef, pork and chicken, respectively. In addition, the results obtained by the portable NIR spectrometer are also good results with R2p prediction coefficient (0.91, 0.82 and 0.83), performance to RPD deviation ratio (3.74, 2.64 and 2.73) and RMSEP of 4.10, 5.81 and 5.61. Finally, our results show that both NIR devices are an alternative for the detection of adulterated alpaca meat.
ABSTRACT This work aimed to train and validate predictive models based on a portable NIR spectrometer and NIR hyperspectral images (NIR-HSI) with chemometric analysis to detect adulteration in alpaca meat. The study was carried out on ground alpaca meat, adulterated with beef, pork and chicken in a range of 0% -50% (p/p). Spectra were obtained using both NIR devices in the spectral region 900 – 1700 nm. Then, chemometric tools such as PCA analysis, SIMCA modeling and PLSR were used to analyze the spectra. PCA captured the spectral variance, so grouping the spectra according to class (alpaca, alpaca + beef, alpaca + pig, alpaca + chicken) was possible. Comparing both devices, it was possible to observe that NIR-HSI better differentiates pure alpaca and adulterated alpaca samples compared to portable NIR. Subsequently, SIMCA was used to create a classification model to distinguish pure alpaca meat (target class) and adulterated alpaca meat. SIMCA results were comparable, with sensitivity and specificity > 95% for both devices. PLSR models were performed using full spectra and selected wavelengths by iPLS. The best results were obtained with the HSI images, with R2p prediction coefficient (0.980, 0.917 and 0.981), yield to RPD deviation ratio (8.39, 3.82 and 7.96) and RMSEP of 1.87, 4.118 and 1.97, for samples of alpaca adulterated with beef, pork and chicken, respectively. In addition, the results obtained by the portable NIR spectrometer are also good results with R2p prediction coefficient (0.91, 0.82 and 0.83), performance to RPD deviation ratio (3.74, 2.64 and 2.73) and RMSEP of 4.10, 5.81 and 5.61. Finally, our results show that both NIR devices are an alternative for the detection of adulterated alpaca meat.
Description
Keywords
Adulteración de alimentos, Carne de alpaca, NIR portátil, Imágenes hiperespectrales, SIMCA