Hyperspectral imaging-based non-destructive detection of freshness changes in MAP stew-braised duck neck during refrigerated storage

Abstract

Stew-braised duck (SBD) products packaged with modified atmosphere packaging (MAP) are prone to quality deterioration during refrigerated storage. Traditional detection methods are time-consuming and invasive. This study aimed to investigate the quality changes of MAP-packaged SBD and to achieve real-time, non-destructive detection using hyperspectral imaging (HSI) without opening the packages. Freshness indicators were evaluated using traditional methods, including pH, total viable count (TVC), low-field nuclear magnetic resonance (LF-NMR), and total volatile basic nitrogen (TVB-N) at 4 °C and 10 °C. A unique image segmentation approach was applied to extract spectral data in the 900–1700 nm range, which were analyzed to evaluate quality changes during 19 days, with a focus on moisture distribution and TVB-N levels. A three-stage fusion strategy involving machine learning models (PLS, RF, PLS-RF), preprocessing techniques (MSC, SG, SNV) and feature extraction methods (CARS, GA, IVSO) was developed. Ultimately, the full-wavelength model at 4 °C using PLS-RF (Rc² = 0.967, RMSEC = 0.710, Rp² = 0.749, RMSEP = 1.951, RPD = 2.026) and the model at 10 °C with SNV-CARS preprocessing using PLS-RF (Rc² = 0.961, RMSEC = 0.944, Rp² = 0.747, RMSEP = 2.431, RPD = 2.003) were identified as optimal for visualizing pixel-level predictions of TVB-N content. This research confirms the feasibility and potential of HSI for non-destructive and rapid detection in MAP-packaged products.