节点文献
畜肉品质评定方法及综合评定系统研究
Study on Quality Assessment Method and Synthesis Evaluation System for Meat of Livestock
【作者】 王笑丹;
【导师】 孙永海;
【作者基本信息】 吉林大学 , 农业机械化工程, 2008, 博士
【摘要】 本论文在吉林省科技发展计划重点项目资助下,对畜肉品质评定方法及综合评定系统进行了研究。建立了畜肉品质综合评定指标体系,确定了评定指标的获取方法。建立了基于色差仪a*值的猪肉颜色评定模型;应用计算机图像处理技术对猪肉大理石花纹等级进行了评定;建立了猪肉品质评定模型。研究开发了基于压力传感器的牛肉嫩度快速评定装置,准确率达到了90%以上,评定时间为5分钟。应用逐步回归法对影响猪胴体瘦肉率的指标及影响牛胴体产肉率的指标进行优选,通过优选后的指标作为参数建立了基于支持向量回归机的猪胴体瘦肉率预测模型和牛胴体产肉率预测模型,核函数选择Poly核函数及RBF核函数构造的混合核函数。应用Matlab及Vc++混合编程技术,开发了畜肉品质综合评定系统,该系统分为4个模块:猪肉品质评定模块、牛肉品质评定模块、猪胴体瘦肉率预测模块和牛胴体产肉率预测模块。
【Abstract】 Along with improvement of living standard, changing of meal structure and extension of international cooperation, the demand of meat of livestock become increasing and the meat of livestock quality become more and more importmant. China have a big margin at various aspects, the main reason is that we still don’t have the meat of livestock system assessment method and standard. It makes the consensus can’t be reached by manufacturer and consumer. The market operation was not standard enough, so it’s difficult to attain a superior quality and excellent price. The local meat of livestock production and foreign trade are influenced by all of these, in order to meet the market requirement, the thesis established a norm meat of livestock quality synthesize assessment system, the method is more objective, quicker and more accurate.The meat of livestock quality synthesize assessment indexes system was established and the indexes getting methods were determined. Eight indexes of pork quality were chose as follow: pH, a* value of lean color, WBSF, TVB-N, cooking loss, MB degree, Dry lean concent and ash concent to assess the pork quality. Most experiment method belongs to nation standard or profession standard. Nine indexes of cutability of pork carcass were chose as follow: carcass weight, carcass length, carcass width of foreside, carcass width of rearward, thickness of back fat, thickness of waist fat, the most thick of fat, fat of triangle meat and rib eye area.The fressness parameter and tenderness parameter of beef were chosed to assess the beef quality. Indexes of retail cut percentages of beef carcass were: gross weight, cold carcass weight, hot carcass weight, bone weight, carcass length, carcass width,thickness of back fat and rib eye area. Hunterlab was used to measure lean color. 1cm fresh pork thickness was chosen as sample thickness from 1cm, 2cm and 2.5cm. L*、a*、b* of lean color were measured, the relativity of sense organs assessment scores and a* value was 0.9230. So we use a* value of hunterlab to predict the lean color and the accuracy of predict equation was above 85%.Image was collected using image collect system which was designed by our group, and using Matlab to assess MB of beef. The relativity of sense organs assessment scores and the image processing technique of computer arrived to 0.9022 and the accuracy of this method was above 85%. The relativity of intramuscular fat content and the image processing technique of computer arrived to 0.8580. Therefore, the image processing technique of computer could predict the MB degree of pork exactly. At last, combine sense organs assessment result and the the intramuscular fat content measurement result to establish MB degree form. The MB content can be predicted by computer image process.Pork quality assessment model was built. The indexes dimension was reduced by PCA. The first fourth PCs were chose as latent vectors of assessment model, and the cumulative contribution rate arrived to 88.37%. Through using multiple linear regression model, BP artificial neural network model and RBF artificial neural network model, establish the best assessment model was established. As a result, multiple linear regression model was chosed as the most accurate model and accuracy was above 85%. So it could predict pork quality by this model exactly and quickly.The inspector of factory can assess pork quality by this method after training.Beef tenderness measuring device was designed. First, the circuits of the device were designed. The FSG-15N1A FS sensor/1500g was chosen and it can output mV signal. Power supply conversion circuit was designed, it could turn 220V AC into±15V AC. Signal amplificatory circuit was designed, the amplificatory times is: 7.28. The relativity of pressure and output voltage arrived to 0.9999. When 1 N weight was put on FS pressure sensor, the output voltage was 136.7 mV. Second, beef tenderness measuring device was designed. The material of sample box was insulating plastic. Choose insulating plastic stick which is 3 cm in length and the 5 mm in diameter as medium of meat sample and testing head of FS sensor. The power supply conversion circuit and signal amplificatory circuit were put in signal process box. The voltage signal was recorded by voltage meter. Third, Parameters of the device was determined. Choose 1.5cm as the experimental depths which the connect stick push into meat samples. The testing time of raw sample was 8-10s after the connect stick was pushed into raw sample. Maximum of output voltage was used as cooked sample testing voltage.The device was validated. Fifty pieces of LD sample were used in this experiment. The relativity of raw meat output voltage and sensory evaluation is 0.9403 and for WBSF is 0.9526. It shows that raw meat output voltage can predict the LD tenderness exactly. The relativity of cooked meat output voltage and sensory evaluation is 0.8704 and for WBSF is 0.8953, it shows that cooked meat output voltage can predict the LD tenderness though it’s a little lower, it’s more complex for cooked meat processing and difficult to collect datum of output voltage exactly by voltage meter. So raw meat output voltage was chose as testing parameter to evaluate the beef tenderness.Beef tenderness assessment equation was established: Y4 = ?2.8313+0.0259VOther 20 LDs was tested by the device and the raw meat output voltage was collected by voltage meter. The result shows that accuracy of the device arrived to 90%. The total testing time of each sample was 5 minutes. This method avoid interfere of man-made factor and the test result was objective, quick and exact.The cutability of pork carcass assessment model was built based on SVR. Choose carcass weight, carcass length, thickness of back fat and rib eye area to predict the cutability of pork carcass by stepwise regress method. The punishing coefficient was 300, the kernel function coefficient was 3 and limiting windage was 0.01. Poly kernel, RBF kernel and mix kernel were chosed as kernel function of SVR to predict the cutability of pork carcass. The accuracy of assessment equation of Poly kernel was 40% and the average windage was 5.77%>5%, it could not predict cutability of pork carcass exactly. The veracity of assessment equation of RBF kernel was 85% and the average windage was 2.93% <5%. The accuracy of assessment equation of mix kernel arrived to 90% and the average windage was 2.99% <5%. Therefore choose mix kernel as the kernel function of SVR.. Comparing SVR model with multiple linear regression model, the accuracy of multiple linear regression model was 85% and average windage was 3.21% < 5%, it could be used to predict the cutability of pork carcass, but it was lower than SVR model. Choose SVR model which based on mix kernel function as assessment model of the cutability of pork carcass.The retail cut percentages of beef carcass assessment model was built based on SVR. Choose cold carcass weight, hot carcass weight, carcass length to predict the cutability of pork carcass by stepwise regress method. The punishing coefficient of SVR was 5, the kernel function coefficient was 0.5 and limiting windage was 0.01. Poly kernel, RBF kernel and mix kernel were chosed as kernel function of SVR to predict the retail cut percentages of beef. The accuracy of assessment equation of Poly kernel was 70% and the average windage was 2.89% <4%, the accuracy of it was low. The accuracy of prediction equation of RBF kernel was 85% and the average windage was 1.88% <4%. The accuracy of assessment equation of mix kernel arrived to 90% and the average windage was 1.85% <4%. Choose mix kernel as the kernel function of SVR. SVR model was compared with multiple linear regression model. The distinct probability of multiple linear regression model was 0.06 >0.05, so the model could not be used. Choose SVR model which based on mix kernel function as assessment model to predict the retail cut percentages of beef carcass.The SVR method was used in cutability of pork carcass assessment and retail cut percentages of beef carcass assessment successfully. To input the datum to program, the prediction result can be attained by computer calculating. It’s simply, quickly, exactly and could be used in factory to inspect on line.The Matlab engine was used to carry out mix programming of Matlab and Vc++. It provided a meat quality synthesis evaluation system which operation surface was friendly and simple. This system included pork quality assessment module, beef quality assessment module, cutability of pork carcass prodiction module and retail cut percentages of beef carcass prodiction module. User could choose the module which he need. This system could implement many tasks such as data process, network training, network testing, data saving etc. And it has many good characters such as simple operating, quick analyses and so on.In summary, the result of assessment meat of livestock quality showed the methods which were chosed could assess the sample exactly, quickly, objectively. And the meat of livestock quality synthesis evaluation system could be used in factory to inspect on line.
【Key words】 Marbling; The cutability of pork carcass; The retail cut percentages of beef carcass; FS sensor; Support Vector Regression;