Studies On The Application Of Computer Vision Technology On The Evaluation Of Grading From Beef Ribeye

As a non-destructive measurement method, computer vision technology has been widely applied in food detection field. However, the research on evaluation of beef grading has not been conducted yet in China. This paper discussed the application of computer vision technology on the evaluation of carcass classification, especially in the color grading and marbling distribution. Furthermore, two programs of grading were developed according to different usage.Based on above goals, this paper finished the following works:The hardware program of computer vision were developed using Visual C ++ 6.0, including Image Analysis 1.0 for image acquisition and morphological features extraction and software of automatic grading for beef ribeye.The usual models of RGB, HIS and Lab system were applied for statistical analysis of sample image which chose randomly. Based on the data analyzed, the distribution regulations of each variable from beef color were obtained, and the best classification parameter T was calculated.This paper also described some algorithms improved by image acquisition methods. By these methods, characteristic parameters of the images were obtained successfully, including Edge examination, gray value transform, median filter, Segmentation of picture, sign area, extract character value, filter. Moreover, some ideal processing parameters were analyzed.In this paper, some correlation models between sensory evaluation and each characteristic parameter, and its accuracy rates were analyzed. Then the best grading models was found and shown as following:The best grading model for fat prediction is FAT = 5.543 - 0.107 * R + 0.140 * G - 0.049 ^*B-0.016 ^*H+0.031 ^*S-0.256 ^*L+0.227 ^*A-0.101 ^*BL+0.261 ^* Eab. Its prediction accuracy rates is 80% as error level lower or equal to 0.3, and 95% as error lever lower or equal to 0.5.The best grading model for muscle prediction is MEAT=4.804+0.046 ^*G-0.073 ^*B +0.001 ^*H-0.094 ^*S+0.212 ^*L+0.136 ^*A+0.009 ^*BL-0.081 ^*Eab. Its prediction accuracy rates is 87.5% as error level lower or equal to 0.3, and 95% as error lever lower or equal to 0.5.The best grading model for marbling prediction is S=5.935+46.165 ^*D1-1427.126 ^* D₁²+6338.641 D₁³. Its prediction accuracy rates is 77.5% as error level lower or equal to 0.3, and 97.5% as error lever lower or equal to 0.5.A typical feed-forward artificial neural network (ANN) were also developed for training and testing samples in this research. The prediction accuracy rates of training samples were 85.7% for fat color, 79.8% for muscle color and 86.7% for marbling. In comparison, The prediction accuracy rates of testing samples were 85% for fat color, 75% for muscle color and 87.5% for marbling.In addition, the software of automatic grading for beef ribeye was programmed. With this software, the prediction accuracy rates was 80%, 70% and 80% for grading of muscle color, fat color and marbling, respectively. At least ten beef ribeye were analyzed in one minute using this software..The objective of this research was to determine the potential of computer vision technology for evaluating beef quality. Results of this research showed that computer system is an effective tool for evaluating meat quality, and would be a substitute of traditional method in future.