Hyperspectral Identification of Composite Chinese Sausages

With the increasing diversity of compound sausage types, the development of a rapid, accurate, and non-destructive identification model has become essential. The aim is to develop a robust model that operates without damaging the sausage surface or structure, meeting real-time, lossless, and high-throughput recognition requirements in practical applications. “Lossless recognition” refers to non-destructive analysis of surface spectra using hyperspectral imaging technology – without chemical treatment, grinding, or slicing into the interior of the sausage. Hyperspectral imaging technology was employed to collect spectral data over 400 – 1000 nm for eight types of compound sausages. For each image, 50 sampling regions were randomly selected, and the average reflectance values were extracted. First, three preprocessing algorithms – Multiplicative Scatter Correction (MSC), Savitzky-Golay smoothing (SG), and Neighborhood Averaging (NA) – were applied to the raw hyperspectral data. To enhance modeling efficiency, Principal Component Analysis (PCA) was used to reduce the dimensionality of the original 328 spectral bands, retaining the first 10 principal components, which explained over 95% of the total variance, as the new feature set. The classification results demonstrate that hyperspectral imaging combined with machine learning algorithms can effectively distinguish between the eight compound sausage types. Among all methods, the SVM model exhibited the highest classification accuracy, highlighting its excellent discriminative ability and robustness in high-dimensional hyperspectral data analysis. Classification models that combined MSC preprocessing with any of the three algorithms achieved prediction accuracies of over 99%.Received30.12.2025Accepted1.03.2026