Effect of PA/PE film thickness on the performance of hyperspectral imaging and machine learning models for predicting residual nitrite in vacuum-packaged pork sausages

초록

Non-invasive prediction of residual nitrite in processed meat products is critical for enhancing food safety and enabling smart manufacturing systems. This study evaluated the feasibility of near-infrared hyperspectral imaging (HSI) combined with machine learning to predict residual nitrite content in emulsified pork sausages vacuum-packaged with polyamide/polyethylene (PA/PE) films of different thicknesses (50-80 mu m). Sausages were formulated with 0, 150, and 300 mg/kg sodium nitrite and analyzed using HSI in the 900-1700 nm range. Distinct spectral differences were observed around 1080 and 1280 nm, which were interpreted as indirect markers of nitrite-associated changes in the sausage matrix, particularly in hydration-related and lipid-associated molecular environments. Three gradient boosting-based ensemble models (CATBoost, LightGBM, and XGBoost) were developed, among which CATBoost achieved the highest predictive accuracy (R2P = 0.969). Notably, variations in PA/PE film thickness did not significantly affect spectral characteristics or model performance within the tested range. Furthermore, pixel-wise prediction and hyperspectral remapping enabled intuitive visualization of the spatial distribution of residual nitrite across sausage samples. These results demonstrate that HSI coupled with an appropriate machine learning algorithm can reliably predict residual nitrite in vacuum-packaged meat products, independent of packaging film thickness. Further validation across diverse product types and processing conditions is required to facilitate industrial implementation of this non-destructive quality assessment approach.

키워드