Flexible Vis/NIR Wireless Sensing and Estimation with DeepEnsemble Learning for Pork

The rapid chilling and aging stages following pork slaughter represent a critical window for determining final physicochemical quality and flavor development. To address the destructive nature of conventional meat quality assessment methods and the limitations of rigid spectral probes when applied to irregular biological surfaces, this study developed and validated a wireless monitoring system integrating a flexible visible/near-infrared (VIS/NIR) sensing array with ensemble learning algorithms. The proposed system enables non-destructive, continuous monitoring of pork quality during cold-chain storage. A DeepEnsemble regression model based on a stacking framework was constructed by integrating Partial Least Squares Regression (PLSR), Support Vector Regression (SVR), and Extreme Gradient Boosting (XGBoost) to predict pH, moisture content, and total amino acid concentration. During a 26 h dynamic aging experiment, the proposed model achieved coefficients of determination (R2) of 0.9019, 0.9687, and 0.9600 for pH, moisture content, and total amino acids, respectively, with prediction performance exceeding that of individual regression models. The wireless transmission module maintained stable data communication under low-temperature and high-humidity conditions (−20 °C and 0–4 °C), with packet loss rates below 0.1%. These results indicate that the proposed system can effectively capture the dynamic evolution of pork quality during aging and provides a practical non-destructive approach for intelligent pork quality evaluation, cold-chain monitoring, and digital management of meat supply chains.