Research On Detecting And Grading Of Pork Quality Based On Computer Vision

Detecting and grading of pork quality are important factors that affect the development of meat industry in China. To establish a uniform system for quality assessment and quality-based pricing of pork are important measures to promote the development of China's meat industry. Using the eye muscle and belly of pork, some fundamental theories and general methods of computer vision applied to pork quality grading were investigated and developed in this work. The main research contents and results were as follows:(1) A computer vision system for pork quality detection was constructed.(2) The background of original RGB image was removed by the method of Maximum Classes Square Error in red component and noise was eliminated by the median filter.(3) In response to the low significant differences of muscle and fat color, the Kernel Fuzzy C-Means Clustering algorithm was used to segment the muscle and fat. And Gaussian kernel function was selected after the comparison of three different kernel functions.(4) An improved watershed algorithm was used to remove the joint of longissimus muscle and surrounding muscle tissue. And comparative analysis of the simple morphological algorithm and traditional watershed segmentation algorithm indicated that the improved watershed algorithm proposed in this paper could extract out longissimus muscle truly and completely with the correct rate of 86.67%.(5) The extracted color features of eye muscle included the mean and standard deviation of R, G, B, H, S, V; marbling features included area ratios of total fat, large fat particles and small fat particles, as well as the number of total fat particles, large fat particles and small fat particles. The correct rates of color and marbling grading were 92.91% and 88.89% respectively.(6) The extracted features of pork belly included stripe number of lean meat, color, uniformity of lean and fat. The color features included the mean and standard deviation of R, G, B; Uniformity features included ratio of lean and fat, variation coefficient of lean area and perimeter, variation coefficient of fat area and perimeter, variation coefficient of total area and total perimeter, ratio of lean and fat area variation coefficient, minimum of lean and fat area variation coefficient. The correct rates of stripe number of lean, color and uniformity were 88.29%, 87.88% and 83.33% respectively.(7) Established a software system for pork quality grading based on VC++.