This study utilized hyperspectral technology in conjunction with chemometric methods for the non-destructive assessment of chilled meat quality. Average spectra were extracted from regions of interest within hyperspectral images and further optimized using seven preprocessing techniques: S-G, SNV, MSC, 1st DER, 2nd DER, S-G combined with SNV, and S-G combined with MSC. These optimized spectra were then incorporated into PLSR and BPNN models to predict TVB-N and TVC. The results demonstrated that the PLSR model employing S-G smoothing in combination with SNV preprocessing yielded optimal predictions for TVB-N (Correlation coefficient = 0.9631), while the integration of S-G smoothing with MSC preprocessing achieved the best prediction for TVC (Correlation coefficient = 0.9601). This methodology presents a robust technical solution for rapid, non-destructive evaluation of chilled meat quality, thereby highlighting the potential of hyperspectral technology for accurate meat quality monitoring through precise quantification of TVB-N and TVC.