Beef Quality Online Grading Technology And System Development

The beef quality score in the world is generally graded by subjective human visual assessment.Beef graders usually determine marbling score according to the fat abundance of beef ribeye region,and finally judge the beef quality score by the reference to the beef muscle color score,fat color score and maturity degree.Using human visual to assessment beef quality is a subjective way which depends on the sense and experience of graders,so it is of applied foreground and research value to develop an objective and efficient beef quality grading system.The embedded technology has its advantages of portability,usability and low costing to develop and automatic beef quality grading system.Using embedded machine vision technology to collect beef image and grading quality score in beef production line can improve the portability,lower the cost and improve work efficiency.So it is of high theoretical value and practical significance to develop beef automatic grading system using embedded machine vision technology.In this paper,machine vision technology and pattern recognition technology are combined,and a beef quality online grading system was developed based on embedded machine vision technology.Full text carried out research around the beef image segmentation and beef quality grading model.Beef ribeye cross images were collected and segmented into different region.Beef marbling features,muscle color features and fat color features were extracted in order to reflect the different grades.Then marbling grading model,color grading model were established according to the features.At last,an beef quality online grading system was developed based on ARM and Windows Mobile handheld device.The main work and conclusions are as follows:1.A beef image segmentation algorithm based on image resampling was introduced.Beef images were firstly converted from RGB color space into Lab color space.Chroma L images were binarized into beef fat region,chroma a images were binarized into beef muscle region.In order to improve the speed of image segmentation to meet the requirements for embedded system,and image binarization method based on image resampling was introduced.Resampling rate of 0.3125 was selected for chroma a image(2592*1552 pixels)binarization,resampling rate of 0.375 was selected for chroma L image binarization.Beef ribeye muscle region was segmented by retaining the largest area using 8-adjacent area labeling method,then ribeye region was segmented by using the same method again.Beef marbling region was segmented by logic XOR operation using beef ribeye region and ribeye muscle region.2.Beef marbling segmentation method based on homomorphic filtering was studied and introduced.Homomorphic filtering was used to correct the non-uniform illumination in the beef ribeye region,and thereby the effects of filtering gain factors and 4 chroma images on marbling segmentation precision were analyzed.Based on this,a beef marbling segmentation method based on hormomorphic filtering with G chroma image was introduced.Experiment results showed that,rL = 0.6 and rH =1.425 were selected to build a homomorphic filter to process the beef ribeye images;the minimum error rate was from G chroma images,which is 5.38%,about 3.73 lower than that without homomorphic filtering.This indicated that with this gain factor,G chroma images after homomorphic filtering can achieve the optimal beef marbling segmentation effect.3.Beef marbling grading model was established.The area of marbling(AM),area of ribeye(AR),number of marbling particles(NM),average area of marbling particles(AAM),area of large marbling particles(ALM),number of large marbling particles(NLM),area of small marbling particles(ASM)and number of small marbling particles(NSM)were first extracted from the processed image,then 7 marbling features were calculated based on these 8 parameters.Linear regress results showed that MF1,MF3,MF4,MF6 and MF7 between beef marbling score were significant.PCA(Principal Component Analysis)was used to shorten these 5 features.A MLR(multiple linear regression)model with accuracy of 100%for marbling grading was established,A BP neural network model with accuracy of 100%for marbling grading was established.4.Beef color grading model was established.The mean and standard deviation of R,G,B,L,a,b were chosen as 12 features to express beef color.Beef muscle color features and fat color features were extracted respectively.Linear regress results showed that all 12 beef muscle color features between beef muscle color score were significant and all 12 beef fat color features except F12 between beef fat color score were significant.PCA was used to shorten 12 muscle color features into 2 principal components and shorten 11 fat color features into 2 principal components.A MLR model with accuracy of 82.61%for muscle color grading was established,A BP neural network model with accuracy of 82.61%for muscle color grading was established.A MLR model with accuracy of 65.22%for fat color grading was established,A BP neural network model with accuracy of 78.26%for fat color grading was established.5.A beef quality online grading system based on ARM core and Windows Mobile OS was developed.Beef image segmentation time for embedded system and the chosen for beef quality grading models were discussed in order to prove the real-time and feasibility of the system.Online experiment showed that,the developed system could get the accuracy of 86.96%for beef marbling grading,the accuracy of 73.91%for beef muscle color grading and the accuracy of 65.22%for beef fat color grading.The operating time using this system between 15-20s was acceptable compare to manual grading time,showing its effectiveness and practicability.