Early detection of beef-quality indicators using hyperspectral imaging combined with pixel-based segmentation method corresponding to fat and protein region

Abstract

This study investigated the utilization of hyperspectral imaging (HSI) in conjunction with pixel-based segmentation to predict the thiobarbituric acid-reactive substances (TBARS) and volatile basic nitrogen (VBN) content in beef. Hyperspectral images were acquired in the visible near-infrared (VIS-NIR) and shortwave infrared (SWIR) ranges to examine temporal alterations in the fat and protein regions. A partial least squares discriminant analysis (PLS-DA) model was employed to segment fat and protein pixels, followed by a partial least squares regression (PLSR) model to predict the TBARS and VBN content from the segmented spectra. The SWIR range yielded the most accurate predictions, with an Rp2 of 0.899 for the early freshness indicators. Utilizing hyperspectral information from individual fat and protein pixels, rather than modeling the entire beef image, resulted in enhanced prediction accuracy for Rp2 of TBARS (0.814–0.899) and VBN (0.394–0.532) in the early stages of storage. These findings elucidate the potential of HSI with pixel-based segmentation as a nondestructive and real-time methodology for precise monitoring of beef freshness.