Prediction of pH in salted mackerel fillets using hyperspectral imaging and chemometrics

• Autores: Arévalo, Diana, Castro, Wilson
• Formato: artículo
• Fecha de Publicación: 2022
• Institución: Universidad Nacional de Frontera
• Repositorio: UNF-Aypate
• Lenguaje: español
• OAI Identifier: oai:ojs2.aypate.revista.unf.edu.pe:article/10
• Enlace del recurso: https://revistas.unf.edu.pe/index.php/aypate/article/view/10
• Nivel de acceso: acceso abierto
• Materia: calidad del pescado; conservación por salazón; perfiles espectrales; aprendizaje automático; fish quality; salting preservation; spectral profiles; machine learning

The objective of this study was to predict the pH of salted mackerel, as a quality indicator, using hyperspectral imaging technology coupled to chemometric techniques. Thirty-five fresh mackerel were acquired in a local market in Sullana, Peru, washed, gutted and filleted to obtain two skinless fillets for each specimen, which were –Ösubjected to a salting process by immersion in 28% brine and stored under refrigeration for 6 days. The pH evaluations and spectra acquisition were carried out with potentiometer and NIR hyperspectral imaging system, respectively on days 0, 1, 2, 3, and 6. The images were corrected, then the sample profiles were extracted by thresholding and pretreated with the Savitzky-Golay filter, followed by implementation of the partial least squares regression (PLSR) model with the full and optimized wavelengths. To validate the model, 30 replicates with cross-validation (K-fold = 5) were applied. The best performance was obtained with PLSR optimized with 9 latent variables, achieving an R2 greater than 0.85 and an RMSE of 0.904. Therefore, the use of HSI NIR with PLSR for pH monitoring in salted fish is feasible.