Machine learning-optimized techniques for enhancing freshness assessment in lamb meat using hyperspectral imaging

Assessment of lamb freshness is crucial to ensure a safe and sustainable lamb supply chain. This study investigates the feasibility of a rapid assessment method for evaluating lamb meat freshness using hyperspectral imaging (HSI). The TVB-N levels of lamb were observed throughout a continuous 12-day refrigeration period, and the variation in TVB-N from day 1 to day 12 was analyzed. Various spectral preprocessing and feature selection techniques were explored to enhance data reliability and reduce dimensionality. Furthermore, four feature wavelength selection methods were integrated with hyperparameter optimization strategies to simplify the model and enhance predictive accuracy. The results indicate that the combination of the whale optimization algorithm with least squares support vector machine based on the variable combination population analysis (VCPA-WOA-LSSVM) yielded superior prediction results (Rp2 = 0.9006, RMSEP = 3.0742, RPD = 2.8464). In addition, the qualitative analysis of lamb meat freshness revealed that the classification of freshness levels achieved 93.33 % accuracy for prediction using the extreme learning machine (ELM) model. The results demonstrate that combining HSI with machine learning-optimized techniques holds significant potential for rapidly and accurately assessing lamb meat freshness.