Visualization accuracy improvement of spectral quantitative analysis for meat adulteration using Gaussian distribution of regression coefficients in hyperspectral imaging

Abstract The percentage of adulterated meat can be hardly recognized by a visual appraisal of their images. In this paper, a novel approach was proposed to combine visualization with pixel discrimination for the purpose of evaluating the adulteration level. Eleven samples (three pure beef samples, three pure chicken samples and five different adulteration level samples) were prepared for hyperspectral imaging (HSI). Each pixel in the hyperspectral images was classified by the Gaussian distribution of regression coefficients (GD-RC) model with three different variables input methods (Each kind of pure samples’ spectra (the unitary GD-RC models) and two kinds of pure samples’ spectra together (the binary GD-RC models) were considered as independent variables, respectively.). Compared with the unitary GD-RC models, the binary GD-RC model had the best performance. The average error (ARE), the correlation coefficient (r), and the root-mean-square error of prediction (RMSEP) of the best method were 2.8%, 0.9831 and 0.0319, respectively. The results demonstrated that the GD-RC model could be used to estimate the adulteration levels of meat samples through the visual appraisal of their images. Therefore, HSI combined with the proposed method can be a powerful and promising tool for meat adulteration visualization.