Quantitative prediction of beef quality using visible and NIR spectroscopy with large data samples under industry conditions

Qiao, T. and Ren, J. and Craigie, C. and Zabalza, J. and Maltin, Ch. and Marshall, S (2015) Quantitative prediction of beef quality using visible and NIR spectroscopy with large data samples under industry conditions. Journal of Applied Spectroscopy, 82 (1). ISSN 0021-9037 (https://doi.org/10.1007/s10812-015-0076-1)

[thumbnail of Qiao-eta-JAS-2015-Quantitative-prediction-of-beef-quality-using-visible-and-NIR-spectroscopy]
Preview
Text. Filename: Qiao_eta_JAS_2015_Quantitative_prediction_of_beef_quality_using_visible_and_NIR_spectroscopy.pdf
Accepted Author Manuscript

Download (1MB)| Preview

Abstract

It is well known that the eating quality of beef has a significant influence on the repurchase behavior of consumers. There are several key factors which affect the perception of quality including color, tenderness, juiciness and flavor. To support consumers repurchase choices, there is a need for an objective measurement of quality that could be applied to meat prior to its sale. Objective approaches such as offered by spectral technologies may be useful, but the analytical algorithms used remain to be optimized. For visible and near infrared (VISNIR) spectroscopy, Partial Least Squares Regression (PLSR) is a widely used technique for meat related quality modeling and prediction. In this paper, a Support Vector Machine (SVM) based machine learning approach is presented to predict beef eating quality traits. Although SVM has been successfully used in various disciplines, it has not been applied extensively in the analysis of meat quality parameters. To this end, the performance of PLSR and SVM as tools for the analysis of meat tenderness is evaluated, using a large dataset acquired under industrial conditions. The spectral dataset was collected using VISNIR spectroscopy with the wavelength ranging from 350nm to 1800nm on 234 beef M. longissimus thoracis steaks from heifers, steers and young bulls. As the dimensionality with the VISNIR data is very high (over 1600 spectral bands), the Principal Component Analysis (PCA) technique was applied for feature extraction and data reduction. The extracted principal components (less than 100) were then used for data modeling and prediction. The prediction results showed that SVM has a greater potential to predict beef eating quality than PLSR, especially for the prediction of tenderness. The influence of animal gender on beef quality prediction was also investigated, and it was found that beef quality traits were predicted most accurately in beef from young bulls.