Quantitative assessment of beef quality with hyperspectral imaging using machine learning techniques

Tools

Ren, Jinchang and Marshall, Stephen and Craigie, Cameron and Maltin, Charlotte (2012) Quantitative assessment of beef quality with hyperspectral imaging using machine learning techniques. In: Hyperspectral Imaging Conference, 2012-05-15 - 2012-05-16.

Full text not available in this repository.Request a copy

Abstract

Nowadays, meat quality assessment attracts increasing attention in hyperspectral imaging as it provides non-intrusive measurement of various attributes of fresh meat such as beef and pork in terms of surface colour, tenderness and pH et al. Typical works can be found in Gowen et al (2007), Maltin et al (2003), Qiao et al (2007) and Eimasry et al (2012), where near-infrared (NIR) hyperspectral imaging systems are usually used. In Qiao et al (2007), selected bands are employed for prediction of drip loss, pH and surface colour of pork, using models determined by neural networks. The prediction correlation coefficients for the three parameters are 0.77, 0.55 and 0.86, respectively. In Eimasry et al (2012), principal component analysis (PCA) is utilised for data reduction, followed by partial least square regression for the prediction of surface colour (L*b*), pH and tenderness, and the prediction correlation values achieved are 0.88, 0.81, 0.73 and 0.83. However, using data captured under commercial conditions in Craigie et al (2010), the prediction correlation values for L*, b*, pH and tenderness are only about 0.2, 0.35, 0.2 and 0.1, respectively. On one hand, this has indicated that the predicted results are critically affected by the conditions when the spectral data are captured. On the other hand, such data is preferred as it reflects commercial conditions for practical applications. Using the challenging data in Craigie et al (2010), support vector machine is applied for prediction of L*, b*, pH and tenderness, followed by dimension reduction using PCA on the entire dataset. With optimised kernels and parameters, we have successfully improved the prediction correlation values to about 0.55, a significant improvement in comparison to the original results.

ORCID iDs

Ren, Jinchang ORCID logoORCID: https://orcid.org/0000-0001-6116-3194, Marshall, Stephen ORCID logoORCID: https://orcid.org/0000-0001-7079-5628, Craigie, Cameron and Maltin, Charlotte;
CORE (COnnecting REpositories)