Development of a Composite Pork Quality Index from Instrumental Traits and Its Hyperspectral Prediction

Pork quality is multidimensional, but conventional evaluation methods are usually destructive, time-consuming, and unsuitable for rapid grading. In this study, longissimus dorsi samples from pigs weighing 30-150 kg were used to construct a composite quality index (CQI) from instrumental traits. Traits related to color, water-holding capacity, antioxidant capacity, and texture were measured, and 13 traits were retained after principal component analysis. The resulting CQI represents a dataset-dependent summary of instrumental quality traits rather than a validated measure of sensory eating quality. Hyperspectral images were collected from pork slices, and CQI was predicted using interval combination optimization (ICO) combined with extreme gradient boosting (XGBoost). In the fixed spectral-level comparison, the SNV-ICO-XGBoost model achieved an R_p² of 0.8670, an RMSE_p of 1.7319, and an RPD_p of 2.7423. Animal-level mean-spectrum validation over 50 random splits showed that PLSR achieved the highest mean prediction performance, with an R_p² of 0.807 ± 0.058, an RMSE_p of 1.857 ± 0.345, and an RPD_p of 2.374 ± 0.345, providing a more conservative assessment of model performance. The predicted CQI values were further visualized qualitatively on the pork surface, supporting further evaluation of HSI for rapid and non-destructive assessment of instrumentally derived pork quality.