Skip to main content
ARS Home » Southeast Area » Athens, Georgia » U.S. National Poultry Research Center » Quality & Safety Assessment Research » Research » Publications at this Location » Publication #345769

Research Project: Assessment and Improvement of Poultry Meat, Egg, and Feed Quality

Location: Quality & Safety Assessment Research

Title: Tenderness classification of fresh broiler breast fillets using visible and near-infrared hyperspectral imaging

Author
item Jiang, Hongzhe - China Agricultural University
item Yoon, Seung-chul
item Zhuang, Hong
item Wang, Wei - China Agricultural University
item Lawrence, Kurt
item Yang, Yi - China Agricultural University
item Jia, Beibei - China Agricultural University

Submitted to: Meat Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/15/2018
Publication Date: 1/31/2018
Citation: Jiang, H., Yoon, S.C., Zhuang, H., Wang, W., Lawrence, K.C., Yang, Y., Jia, B. 2018. Tenderness classification of fresh broiler breast fillets using visible and near-infrared hyperspectral imaging. Meat Science. 139:82-90.

Interpretive Summary: Boneless skinless breast meat is the most popular chicken meat product in North American and many European countries. Tenderness is an important quality attribute and plays an important role in eating quality of chicken breast meat. Although numerous techniques have been used to predict or measure chicken breast meat tenderness and provide reliable information, these methods are time-consuming and invasive. Hyperspectral imaging (HSI) is a novel analytical technology which could provide both spectral and spatial information of tested food samples. Many studies have shown that HSI can be successfully used to predict quality of various food products. In the poultry industry, the HSI technique has also been applied to assess the physical and chemical attributes of chicken meat, and the results are promising. However, so far few studies have been conducted to evaluate the tenderness of chicken meat employing HSI. Therefore, the objective of the present study was to test the potential of Vis/NIR (400-1000 nm) HSI for classifying chicken breast meat based on their tenderness. Results suggest that the HIS with full spectral data can classify the fillets into three categories (i.e. tough, tender, very tender) compared to corresponding instrumental shear force values. Furthermore, using specifically selected wavelengths data and image textural features improves results for the prediction. The distribution map based on data analyses successfully showed visual differences in tenderness between different chicken breast fillets and between the regions within an individual fillet. These results suggest that HSI technique could be used to sort chicken breast meat based on its tenderness on line non-destructively and rapidly.

Technical Abstract: The potential for visible and near-infrared (Vis/NIR, 400-1000 nm) hyperspectral imaging (HSI) to classify intact fresh broiler breast fillets (pectoralis major) and visualize fillet tenderness was assessed in this work. A total of 75 broiler breast fillets were collected from a commercial broiler processing plant and hyperspectral images of the fillets were acquired. Warner-Bratzler shear force (WBSF) values of 150 samples (two strips for each fillet) were collected as an indicator of meat tenderness and accordingly all fillets were grouped into three categories: tough, tender and very tender. Spectra were extracted from the region of interests (ROIs) manually selected corresponding to positions of WBSF measurements, followed by an explanatory principal component analysis (PCA) transform of images. Image textural data of ROIs were extracted on the first three principal component (PC) images using grey level co-occurrence matrix (GLCM) method. Firstly, based on full wavelengths, the classification model was explored using partial least square discriminant analysis (PLS-DA) and radial basis function-support vector machine (RBF-SVM) with various preprocessing methods, and the best correct classification rate (CCR) was 0.90 for the prediction set using PLS-DA and optimal preprocessing of second derivative (Der2). Furthermore, as a comparison, 15 tenderness-related wavelengths were selected using uninformative variable elimination (UVE) and successive projections algorithm (SPA) to build a new model with a lower CCR value (0.70) but more practical to implement. Combining spectra of selected/partial wavelengths with image textural data increased CCR values of the model, which indicates that image textural data benefit the predicting model. Furthermore, in order to visualize of the meat with different tenderness, fillet images were masked by bands after the influences of background and specular reflection were removed. Then distribution (or prediction) maps were created with each pixel that was classified based on meat tenderness in the masked hyperspectral images and showed that the chicken breast fillets from different tenderness categories had readily discernible images. In conclusion, the Vis/NIR HSI technique is a feasible and non-destructive methodology for predicting the tenderness of intact fresh broiler breast meat, although its performance and practicality remain to be further validated.