Rapid determination of lamb meat freshness using the hyperspectral imaging combined with symmetric stacking ensemble algorithm

Abstract

Freshness is a key indicator in determining the quality of lamb meat. This study explores the feasibility of a rapid detection method for assessing lamb meat freshness through hyperspectral imaging. The variations in total volatile basic nitrogen (TVB-N), total viable count (TVC), pH, and lightness (L*) of lamb samples were analyzed over 12 days. Variable combination population analysis (VCPA) was employed to enhance data reliability and reduce dimensionality, while a symmetric stacking ensemble learning (SSEL) network was developed to predict both the freshness indices and the storage duration of lamb meat. Consequently, the feature wavelengths for each freshness index were identified. In particular, the spectral peak in the 620–630 nm range emerged as a crucial biomarker wavelength for evaluating lamb meat freshness during storage. The results demonstrate that the SSEL network outperforms the optimal traditional model for each indicator. Specifically, the SVM-stacking model exhibits outstanding performance for TVB-N (R_p² = 0.93, RMSEP = 2.28), while the random forest (RF) stacking model excels in predicting TVC (R_p² = 0.91, RMSEP = 0.84), pH (R_p² = 0.89, RMSEP = 0.19) and L* (R_p² = 0.88, RMSEP = 1.83). In addition, the SVM-stacking model also surpassed traditional approaches in predicting the storage duration of lamb meat, with R_p² and RMSEP values of 0.93 and 2.28, respectively. The proposed methodology enables rapid freshness evaluation and captures temporal variability while offering insights into the molecular mechanisms behind spectral variations. This research lays a foundation for the accurate detection of meat product quality.