针对牛肉大理石纹人工评级过程中人为误差干扰大的问题，研究利用图像处理技术提高牛肉大理石纹评级的客观性并增强自动化程度，提出基于不变矩、灰度共生矩阵和混沌蜂群优化混合核函数支持向量机（Support Vector Machine，SVM）的牛肉大理石纹评级法。首先计算牛肉大理石纹图像的不变矩和灰度共生矩阵统计量并由此构造特征向量；然后将训练和测试样本输入到混合核函数SVM，采用混沌蜂群算法优化SVM中的惩罚因子及核参数，使其分类识别性能达到最优；最后输入待评级样本进行分类识别，得到最优评级结果。大量实验结果表明：根据我国NY/T676-2010牛肉大理石纹标准图谱，评级正确率分别达到100%（一级）、93.3%（二级）、93.3%（三级）、96.7%（四级）、100%（五级）。与基于灰度矩和SVM法、基于灰度共生矩阵和BP（Back Propagation）神经网络法相比，本文所得评级正确率最高，且与专业评级师的实际评级情况最相符。

The image processing techniques was used to improve the objectivity and degree of automation in the grading of beef marbling, in order to minimize the interference caused by human errors in the manual beef marbling grading process. This study proposed the use of a grading method for beef marbling that utilized the invariant moments, a gray level co-occurrence matrix, and a mixed kernel support vector machine (SVM), optimized by a chaotic bee colony. Firstly, the invariant moments and the statistical quantity of gray level co-occurrence matrix of the beef marbling image were computed in order to construct a feature vector. The training and testing samples of the beef marbling image were then inputted to a mixed kernel function SVM. Optimal recognition performance was attained by optimizing the penalty factor and kernel parameters of the mixed kernel function SVM using a chaotic bee colony algorithm. Finally, the samples to be graded were inputted to the SVM for classification and recognition, and the optimal grading results were obtained. A large number of experimental results revealed grading accuracies of 100% (Grade One), 93.3% (Grade Two), 93.3% (Grade Three), 96.7% (Grade Four), and 100% (Grade Five), based on the standard beef marbling image obtained by NY/T676-2010. The proposed method showed the highest grading accuracy compared to those of the methods developed utilizing gray moment and SVM, and the gray level co-occurrence matrix and the black propagation (BP) neural network; in addition, the obtained results were closest to the actual grading results obtained by the professional beef marbling grading division.

国家自然科学基金资助项目(60872065)；江南大学食品科学与技术国家重点实验室开放基金项目(SKLF-KF-201310)；江苏高校优势学科建设工程资助项目