Potential of hyperspectral imaging and multivariate analysis for rapid and non-invasive detection of gelatin adulteration in prawn

In this study, the reliability and accuracy of hyperspectral imaging was investigated for detection of gelatin adulteration in prawn. The spectra of prawns were extracted according to the shape information of prawns contained in the hyperspectral images. Least-squares support vector machines (LS-SVM) was used to calibrate the gelatin concentrations of prawn samples with their corresponding spectral data. The combination of uninformation variable elimination (UVE) and successive projections algorithm (SPA) was applied for the first time to select the optimal wavelengths in the hyperspectral image analysis. The UVE–SPA–LS-SVM model led to a coefficient of determination (rP2) of 0.965 and was transferred to every pixel in the image for visualizing gelatin in all portions of the prawn. The results demonstrate that hyperspectral imaging has a great potential for detection of gelatin adulteration in prawn.