Abstract: Physiological maturity is an important indicator to determine the quality grade of beef. This paper proposes a method to predict the physiological maturity of beef by using a support vector machine (SVM) model with parameters optimized by an improved grid search (IGS) algorithm. A total of 100 beef samples at different slaughter ages of 18, 36, 54 and 72 months (25 for each age) were collected. Using machine vision, the microscopic images of the samples were collected. After image processing, the characteristic parameters of muscle fibers from beef with different physiological maturity were extracted, and the correlation between the physiological maturity of beef and the characteristic parameters of muscle fibers was analyzed by statistical methods. Using the characteristic parameters of muscle fibers as the input, a training set of 76 samples was used to establish a SVM prediction model for beef physiological maturity. An improved grid search algorithm was proposed to optimize the constraint parameter C and the kernel function parameter g of the SVM model. Furthermore, using the leave-one-out cross validation method, the optimal parameter combination (C, g) was obtained and substituted into the classifier to obtain an optimized prediction model for beef physiological maturity. The applicability and estimation performance of the prediction model were tested with independent samples from 24 test sets. The results showed that the accuracy of the prediction model was up to 91.67%. Compared with the traditional grid search algorithm, the IGS algorithm could reduce the model training time by 1 755.41 s. There was a significant correlation between beef muscle fiber characteristics and slaughter age. According to the characteristic parameters of beef muscle fiber, the physiological maturity of beef could be determined automatically using machine vision technology.

Key words: beef, physiological maturity, support vector machine, prediction model