Nondestructive evaluation of beef palatability

Scope and method of study. The U.S. beef industry currently uses the USDA beef quality grading as a marketing tool to categorize carcasses on the basis of palatability. This manual grading system is subjective. In this study, a computer vision system was developed to predict USDA quality grade, objectively. An adaptive segmentation algorithm was developed to separate the longissimus dorsi (l.d.) muscle from the ribeye in a digital image. Based on color and marbling distribution in the l.d. muscle, a regression model predicted USDA quality grade. A bioequivalence statistical analysis method was implemented to determine if the grades predicted by the computer vision system were significantly equivalent to the grades assigned by expert graders.; Consumers consider tenderness a primary determinant of beef palatability. A direct measurement of beef tenderness is not considered in USDA beef quality grading, because there is no suitable nondestructive method available to the beef industry. This study examined three nondestructive methods for predicting aged, cooked-beef tenderness; computer vision, near-infrared (NIR) spectroscopy, and X-ray imaging. Textural features extracted from images of fresh beef using statistical methods, Gabor filters, and wavelets were used to predict tenderness. A portable NIR spectrometer was developed and tested online in beef packing plants. Transmitted X-ray images of beef steaks were acquired at various energy levels. A dual-energy X-ray absorption model was developed to predict tenderness.; Findings and conclusions. The adaptive segmentation algorithm separated l.d. muscle with 98% accuracy. Grades predicted by the computer vision system were statistically equivalent to the grades assigned by expert graders. A linear regression model using statistical textural features predicted shear-force tenderness with an R2 value of 0.72. A canonical discriminant model using Gabor textural features classified carcasses into three tenderness groups with 79% accuracy. A stepwise regression model using wavelet textural features successfully classified carcasses into nine tenderness certification levels. A partial least squares model was developed to sort beef carcasses based on NIR spectral scans of the ribeye surface. At up to 70% certification levels, the system successfully sorted tough from tender carcasses. If 30% of the tough carcasses were removed, the remaining 70% of the carcasses could be marketed as "guaranteed tender." X-ray attenuation properties were found to be unrelated to tenderness.