Visible/Near-Infrared Hyperspectral Imaging for Beef Tenderness Prediction

Authors

Govindarajan Konda Naganathan, University of Nebraska-LincolnFollow
Lauren M. Grimes, University of Nebraska-Lincoln
Jeyamkondan Subbiah, University of Nebraska-LincolnFollow
Chris R. Calkins, University of Nebraska-LincolnFollow
Ashok Samal, University of Nebraska-LincolnFollow
George E. Meyer, University of Nebraska-LincolnFollow

Date of this Version

2008

Comments

Published in Computers and Electronics in Agriculture 64 (2008): 225-233; doi: 10.1016/j.compag.2008.05.020 Copyright © 2008 Elsevier Ltd. Used by permission. http://www.sciencedirect.comjournalhomepage:www.elsevier.com/locate/compag

Abstract

Beef tenderness is an important quality attribute for consumer satisfaction. The current beef quality grading system does not incorporate a direct measure of tenderness because there is currently no accurate, rapid, nondestructive method for predicting tenderness available to the beef industry. The objective of this study was to develop and test a visible/nearinfrared hyperspectral imaging system to predict tenderness of 14-day aged, cooked beef from hyperspectral images of fresh ribeye steaks acquired at 14-day post-mortem. A push-broom hyperspectral imaging system (wavelength range: 400-1000 nm) with a diffuse-flood lighting system was developed and calibrated. Hyperspectral images of beef-steak (n = 111) at 14-day post-mortem were acquired. After imaging, steaks were cooked and slice shear force (SSF) values were collected as a tenderness reference. All images were corrected for reflectance. After reflectance calibration, a region-of-interest (ROI) of 200 × 600 pixels at the center was selected and principal component analysis was carried out on the ROI images to reduce the dimension along the spectral axis. The first five principal components explained over 90% of the variance of all spectral bands in the image. Gray-level textural co-occurrence matrix analysis was conducted to extract second- order statistical textural features from the principal component images. These features were then used in a canonical discriminant model to predict three beef tenderness categories, namely tender (SSF ≤ 205.80 N), intermediate (205.80 N < SSF < 254.80 N), and tough (SSF ≥ 254.80 N). With a leave-one-out cross-validation procedure, the model predicted the three tenderness categories with a 96.4% accuracy. All of the tough samples were correctly identified. Our results indicate that hyperspectral imaging has considerable promise for predicting beef tenderness.