Hyperspectral imaging combined with chemometric analysis for non-destructive prediction of mutton quality during refrigerated storage

During refrigerated storage (4℃), mutton quality parameters (TVB-N, TVC, pH and the color L *-value) and freshness change significantly, serving as key indicators. Conventional assessment methods are typically destructive, time-consuming and lack spatial mapping capability. This study investigates hyperspectral imaging (HSI) combined with chemometrics for the rapid, non-destructive prediction of mutton quality parameters and visualization of their distribution maps. The results indicated that the CARS-PLSR model for TVB-N prediction (RPD = 3.6432), the RFLA-PLSR model for TVC prediction (RPD = 3.2150), the CARS-ELM model for pH prediction (RPD = 3.2451), and the CARS-ELM model for the color L *-value prediction (RPD = 3.3367) exhibited the best performance. Subsequently, the distributions maps of TVB-N, TVC, pH and the color L *-value in mutton were generated using their respective optimal prediction models were visualized. Among the tested classification approaches for freshness evaluation, Multinomial Logistic Regression (MLogR) discrimination models built using CARS exhibited higher accuracy (98.29 %) for the calibration set and highest discrimination accuracy (98.87 %) for the prediction set. These results confirm the potential of hyperspectral imaging as an effective tool for rapid and non-destructive assessment of mutton quality and freshness. • Reliable prediction models for mutton quality parameters (TVB-N, TVC, pH, the color L* - value). • Optimal spectra preprocessing identified from five candidate methods. • Employed CARS and RFLA to select feature wavelengths. • Achieved accurate freshness classification using PLS-DA and MLogR models.