Evaluation of broiler breast fillets with the woody breast condition using expressible fluid measurement combined with deep learning algorithm

Authors: YANG, YI - China Agricultural University; WANG, WEI - China Agricultural University; Zhuang, Hong; Yoon, Seung-Chul; Bowker, Brian; JIANG, HONGZHE - Nanjing Forestry University; PANG, BIN - Qingdao Agricultural University

Submitted to: Journal of Food Engineering
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 5/8/2020
Publication Date: 5/13/2020
Citation: Yang, Y., Wang, W., Zhuang, H., Yoon, S.C., Bowker, B.C., Jiang, H., Pang, B. 2021. Evaluation of broiler breast fillets with the woody breast condition using expressible fluid measurement combined with deep learning algorithm. Journal of Food Engineering Volume 288, 110133. https://doi.org/10.1016/j.jfoodeng.2020.110133.
DOI: https://doi.org/10.1016/j.jfoodeng.2020.110133

Interpretive Summary: Water-holding capacity (WHC) is defined as the ability of muscle to retain its inherent or added water. It is one of the critical quality attributes of meat. Many studies have investigated the effects of the Wooden Breast (WB) condition on WHC indicators such as cook loss and drip loss and have shown that the WB condition of meat results in significantly reduced WHC. Expressible Fluid (EF) is a commonly-used WHC indicator. Its results indicate total free or loosely bound water in meat. However, there is a lack of data showing the relationship between WHC measured with the EF method and the woody breast condition in broiler breast fillets. Deep learning (DL) is a new machine learning algorithm and has shown promising results in many research fields. Compared with conventional machine learning methods, the advantage of deep learning (DL) include: 1) it can process larger datasets; 2) it uses raw or unprocessed data or images in evaluations; and 3) it can extract deep features or information, which is not easily recognized by conventional methods. The objective of the present study was to evaluate the DL method to determine the relationship between the EF measurements (or WHC measured with the EF method) and the WB condition in broiler breast fillets. Our results show that there were significant differences in average EF measurements between normal and WB meat. The DL method further demonstrates that there was a close relationship between the WB severity and WHC in broiler breast fillets based on EF images. Data suggest that the WB condition significantly affects meat WHC measured by the EF method. The DL method might be used to predict WHC of meat based on the EF images.

Technical Abstract: In this study, the relationship between expressible fluid (EF) measurements and the woody breast (WB) condition in broiler breast fillets (pectoralis major) was investigated and the deep learning algorithm (DLA) technique was evaluated to predict degrees of the WB condition based on EF images. Fillet samples were collected from a commercial plant and categorized into normal (no WB), moderate WB, and severe WB groups. Expressible fluid of fresh and frozen samples was measured using the filter paper press method. The features of the images were analyzed using traditional manual method, gray level co-occurrence matrix (GLCM), method and the DLA respectively. The results show that there were significant differences in average EF measurements between three WB categories (P < 0.05) regardless of fillet state (Fresh or Frozen). The DLA feature, instead of EF ratios, showed a close relationship between the WB grade and WHC in broiler breast fillets directly based on EF images. The accuracy in classification of WB grades could be as high as 93.3% for fresh and 92.3% for frozen fillets in independent validation set. Data suggest that the WB condition significantly affects the meat WHC measured by the EF method. The deep learning algorithm provides a useful reference for the assessment of the EF images.