为了解决牛肉大理石花纹在人工分级中准确率和效率低的问题，该文基于计算机视觉和图像处理技术提出一种实用的牛肉大理石花纹自动评估和分级方法。通过图像解析，利用所提出的分级算法实现牛肉大理石花纹的快速提取，并计算反映大理石花纹丰富程度的10个特征参数。使用特征参数建立主成分回归模型，对牛肉大理石花纹等级进行预测，预测相关系数 Rv=0.88，预测标准差 SEP=0.56。校正时模型总体的回判正确率为97.0%，验证时总体的判别正确率为91.2%。在此基础上，开发了大理石花纹自动分级软件系统和硬件装置，并且在该试验室前期研制的样机上进行了试验，验证了算法的运算速度和准确率。结果表明，所提出的分级方法的检测速度和精度均能够满足企业中对牛肉大理石花纹分级的要求。%Beef marbling is one of the most important indexes to assess beef quality. The grade of beef marbling is a measure of the fat distribution density in the rib-eye region. However, quality grades of beef in most of the beef slaughtering houses and enterprises depends on trainees using their visual senses or comparing to the standard sample cards in China. This manual grading method demands not only great labor but also lacks objectivity and accuracy. The objective of this research was to investigate an optimal method for grading the beef marbling based on computer vision and image processing technologies to meet the requirement of the meat industry. A practical algorithm that can be used in a beef marbling grading system is proposed in this research. The beef sample images were collected by a machine vision image acquisition system. The system consisted of an image acquisition device, computer, and image processing algorithm equipped into the self developed system software. The images of the beef samples in an aluminum plate were captured by CCD. Light intensity was regulated through a light controller, and the distance between the camera lens and the beef samples was adjusted though translation stages in the image acquisition device. Collected images were automatically stored in the computer for further image processing. First, some methods such as image denoising, background removal, and image enhancement were adopted to preprocess the image to obtain a region of interest (ROI). In this step, the image was cropped to separate the beef from the background. Then, an iteration method was used to segment the beef area, obtain the beef marbling area and fat area. The redundant fat area was removed to extract an effective rib-eye region. Ten characteristic parameters of beef marbling namely, the rate of marbling area in the rib-eye region, the number of large grain fat, medium grain fat, small grain fat, total grain fat, the density of large grain fat, medium grain fat, small grain fat, total grain fat and, the evenness degree of fat distribution in the rib-eye region can reflect the amount of marbling and its distribution. So they were used to establish principal component regression (PCR) model. The PCR model result yielded a correction coefficient (Rv) of 0.88 and a standard error of prediction (SEP) of 0.56. And the PCR model showed that the rate of the marbling area in the rib-eye region had the greatest effect on the grade of beef marbling. Fisher discriminant functions were constructed based on the PCR model results to classify the grade of beef marbling. Experimental results showed that the classification accuracy was 97.0%in the calibration set and 91.2%in the prediction set. On this basis, a software system was developed for the automatic grading of beef marbling. A corresponding hardware device was also developed, controlled by the software system for real time application. The speed and accuracy of the algorithm were verified with theoretical analysis and a practical test. Through tests, the average recognition time of each sample was 0.879 s. The results showed that the algorithm could meet the beef marbling testing and grading in precision to the practical application. Moreover, this method is of great significance for the development of an automatic classification system.
展开▼
