Quantification and visualization of meat quality traits in pork using hyperspectral imaging. (February 2023)

• Record Type: Journal Article
• Authors: Tang, Xi; Rao, Lin; Xie, Lei; Yan, Min; Chen, Zuoquan; Liu, Siyi; Chen, Liqing; Xiao, Shijun; Ding, Nengshui; Zhang, Zhiyan; Huang, Lusheng
• Is Part Of: Meat science. Volume 196(2023)
• Publication Date: 2023-02
• Subjects: Hyperspectral imaging -- VIS/NIRS -- Meat quality -- Pigs -- Visual appraisal; ANN Artificial neural network -- AOAC Association of Official Analytical Chemists -- GLCM Gray level co-occurrence matrix -- GWAS Genome-wide association analysis -- HSI Hyperspectral imaging system -- R2cal Determination coefficients of calibration -- R2cv Determination coefficients of cross-validation -- RMSEC Root mean square error of calibration -- RMSECV Root mean square error of cross validation -- ROI Region of interest -- RPD Residual predictive deviation -- SWAS Spectral-wide association analysis -- VIS/NIRS Visible/near-infrared spectroscopy; Meat -- Periodicals; Meat industry and trade -- Periodicals; Viande -- Périodiques; Viande -- Industrie -- Périodiques; Meat; Meat industry and trade; Periodicals; 641.36
• DOI: 10.1016/j.meatsci.2022.109052
• Languages: English
• ISSNs: 0309-1740

Abstract

Abstract: Accurate and rapid determination of meat quality traits plays key roles in food industry and pig breeding. Currently, most of the spectroscopic instruments developed for meat quality determination can only obtain the spectral average value of the sample, so it is difficult to evaluate the spatial variation of meat quality traits. In this study, we evaluated the predictive potential of 14 meat quality traits based on large-scale VIS/NIR hyperspectral images collected by SpecimIQ. When predictions were based solely on hyperspectral data, the prediction accuracy (R 2 cv ) for the majority of meat qualities ranged from 0.60 to 0.70. After adding texture information, the prediction accuracy of all traits is improved by different magnitudes (R 2 cv increases from 1.5% to 16.4%). Finally, the best model was utilized to visualize the spatial distribution of Fat (%) and Moisture (%) to assess their homogeneity. These results suggest that hyperspectral imaging has great potential for predicting and visualizing various meat qualities, as well as industrial applications for automated measurements. Highlights: This is a meat measurement process more suited to an assembly line operation. Extraction of texture feature from hyperspectral images based on GLCM. Use SWAS to validate the outcomes of artificial neural network model's prediction. Spectral information combined with texture features improves prediction accuracy. Visualize the distribution of Fat(%) and Moisture(%) in the longissimus muscle.