Simultaneous detection and visualization of lipid and protein oxidation in frozen-thawed chicken meat using hyperspectral imaging.

The quality degradation of frozen meat during storage is mainly attributed to oxidative reactions in lipids and proteins. This study aimed to evaluate lipid and protein oxidation in frozen-thawed chicken meat by employing hyperspectral imaging (HSI). During ten freeze-thaw (F-T) cycles, the thiobarbituric acid reactive substances (TBARS) content in chicken meat increased from 0.0910 to 0.5120 mg/kg, and the carbonyl content increased from 1.0560 to 3.9550 nmol/mg. Correlation analysis result indicated a strong correlation (r = 0.982) between TBARS and carbonyl content. To address the inefficiency of conventional methods that require training separate models for each indicator, a novel multi-task deep learning framework integrating gramian angular difference fields (GADF) with multi-task convolutional neural network (MTCNN) was proposed to achieve end-to-end simultaneous prediction of TBARS and carbonyl content in a computer vision processing paradigm. The developed GADF-MTCNN model demonstrated superior predictive performance for correlated tasks, significantly outperforming the compared single-output deep learning models, with prediction results of R2p = 0.9458, RMSEP = 0.0296 mg/kg, and RPD = 4.3030 for TBARS, and R2p = 0.9545, RMSEP = 0.1905 nmol/mg, and RPD = 4.7014 for carbonyl content. Moreover, lipid and protein oxidation were simultaneously visualized in a pixel-wise manner by transferring the established GADF-MTCNN model to hyperspectral images, which contributes to the management of storage and sales. The results indicated the combination of HSI and GADF-MTCNN has great potential for evaluating lipid and protein oxidation in frozen-thawed chicken meat.