Classification of fish meal produced in China and Peru by online near infrared spectroscopy with characteristic wavelength variables
Online near infrared reflectance spectroscopy combined with characteristic wavelength variables was used to establish a fast and nondestructive analytical method for the classification of fish meal produced in China and Peru. In this study, 117 fish meal samples (47 from China and 70 from Peru) were scanned in the spectral range of 1000–2500 nm by the online near infrared spectroscopy instrument applied on the conveyor belt. The K–S (Kennard–Stone) method was used for the division of samples into calibration and validation sets. Principal component analysis and partial least square discriminant analysis were applied to classify fish meal samples. The results showed that the discrimination accuracies with calibration and validation set samples were 100% and 89.74%, respectively, for the partial least square discriminant analysis model using the full spectrum after the optimimal spectral pre-treatment. Then competitive adaptive reweighted sampling (CARS) was used to select the characteristic wavelength variables for partial least square discriminant analysis model analysis, and the discrimination accuracy for the validation set increased to 94.87%. All the results indicated that online near infrared spectroscopy combined with characteristic wavelength variables could be used for discriminating fish meal samples produced in different places, which offers feed purchasers an effective, reliable, and real-time analysis method for the identification and authentication of the commercial fish meal product.