| The pork carcass grading is an important factor to influence the commercial value of raisers, slaughterhouses and meat enterprises. In market, pig weight and lean yield influence the commercial value. To increase commercial values of meat business, we must objectively grade pork carcass in the slaughtering line. In this paper, the computer vision is used to grade pork carcass. Combining with ready-made method and knowledge, we employed Artificial Neural Network(ANN), Wavelet, Fuzzy Theory, Regression Analysis, Image Processing, Software technology and so on. To complete the research of system of pork carcass grading, we had done much work and obtained some conclusions:1. We summarized and analyzed the studies about pork carcass grading, pork color, lean yield, marbling and tenderness. We discussed the advantages and disadvantages of the methods at present.2. Principal component analysis. We analyzed lean yield features, and found the relation between them.3. The study of lean yield estimation. The traditional method of lean yield estimation not only wastes money and time, but also affects the accuracy of result because of incompletely exfoliation. Except these, it lowers the pork price and causes economic loss because it is a destructive experiment, so we should find a reasonable method. In this paper, we used carcass feature and other features as important input, and multiple linear regression (MLR) was used to predict lean yield. We concluded the function:We tested the function with 30 samples. The mean error is 3.41% and RSD is 0.061, they are all full with conditions. The average value of prediction and the estimated value is proved to have no significant difference through the T-test (P<0.05). 4. Predicting the color score of muscle. We used MLR and ANN to predict the color score of muscle. The MLR function:We tested the function with 10 samples, and the accurate value is 80% and 70%. The result of ANN is satisfied. 5. Predicting marbling score . The marbling affects pork quality and is one of norms to grade carcass, so it is very important to correctly estimate it. We used LR, MLR and ANN to predict the score. The LR function:The MLR function:Through testing, the accurate value is 75%, 87.5% and 87.5%. So the method of MLR and ANN is feasible.6. Predicting tenderness score. We discussed the theory and the study of method of obtaining the texture feature. In the paper, we studied the contribution of Co-occurrence Matrix which could demarcate the texture feature. The tenderness is an important factor in consumers' perception of pork quality. We used Co-occurrence Matrix, Method of combining Co-occurrence Matrix and Energy to predict the tenderness score. In the method of Co-occurrence Matrix, we used MLR and ANN to predict the score, their accurate values respectively are 66.67% and 83.33%. In the method of Co-occurrence Matrix and Energy, we used MLR to predict the tenderness score, its accurate value is 66.7%. Compared two methods, the Co-occurrence Matrix is more accurately7. Grading pork carcass and Last-Description (L-D). In the paper we used carcass feature, color score of muscle, marbling score, tenderness score, lean yield and carcass weight to predict pork carcass grade. We used MLR and ANN to predict pork carcass grading .The MLR function:their accurate values are 80%and 90%. We combined marbling, tenderness, muscle color with carcass grading to complete carcass L-D.8. Designing system. Through image processing, we designed the grading system. Through the test, the system could predict the score of muscle color, marbling, tenderness and grade accurately. It needs less time and money than the former methods.9. Developing system. The system software is developed with C++ program which is OO. With the tool of VC, we make the software of image processing and pattern recognition. It is effective and practical. Based on the software we develop an auto system software for slaughterhouses and detection. The user can make the system with clavier and mouse conve...
|
PDF Full Text Request