Location: Quality and Safety Assessment Research UnitTitle: Prediction of quality attributes of chicken breast fillets by using hyperspectral imaging technique combined with deep learning algorithm
|YANG, YI - China Agricultural University|
|WANG, WEI - China Agricultural University|
|JIANG, HONGZHE - China Agricultural University|
Submitted to: ASABE Annual International Meeting
Publication Type: Proceedings
Publication Acceptance Date: 5/1/2019
Publication Date: 7/17/2019
Citation: Yang, Y., Wang, W., Zhuang, H., Yoon, S.C., Bowker, B.C., Jiang, H. 2019. Prediction of quality attributes of chicken breast fillets by using hyperspectral imaging technique combined with deep learning algorithm. ASABE Annual International Meeting. Paper No. 1901301. https://doi.org/10.13031/aim.201901301.
Interpretive Summary: As chicken meat is typically sold by weight, the ability of water retention by meat, or water holding capacity (WHC) of meat, becomes an important property. Among several WHC indicators, drip loss is most commonly used to predict losses of fresh meat during storage and quality in displays. We investigated the potential of deep learning and near-infrared (NIR) hyperspectral imaging in the spectral range of 1,000 and 2,500 nm to predict drip loss categories of chicken breast fillets. Stacked auto-encoders and softmax regression were used to extract deep features and establish a classification model. The results showed that deep learning features had a better predictive ability than features from either full or key wavelengths. The deep learning method achieved the accuracy of 90% in classifying two drip loss groups.
Technical Abstract: Hyperspectral imaging (HSI) with a wavelength range between 1000 and 2500 nm was evaluated as a potential technique to grade chicken breast fillets based on drip loss. The deep learning, as an effective method, was also evaluated in extracting the feature variables from hyperspectral data. Both full and selected key wavelengths of spectra were used in establishing partial least-squares discriminant analysis (PLS-DA) model. Stacked auto-encoders and a Softmax regression (SAEs-Softmax) model were used to extract deep feature variables and establish a classification model based on the deep learning algorithm. Results showed that deep learning features had better predictive ability compared with the feature variables of either full or key wavelength. The classification accuracies of SAEs-Softmax were 90% for two drip loss groups. Results revealed that the deep learning method may enhance quality assessment of chicken meat by HSI technique.