Research On Pig Fat Content Base On Ultrasound Image

With the development of China’s national standard of living,pork quality requirementsare constantly improve.There are many factors that influence the quality of pork, and have avariety of conditions to limit in the process of detection. How to realize low cost, rapid,reliable of detection for pig meat has important significance. This paper studies based on porkloin Ultrasonography Images, realize the detection and recognition of its fat content.Pork loin commonly known as pork, intramuscular fat content is an important referenceindex to reflect the overall fat content for a pig. Because the pig fat content can reflect theoverall situation of pig meat quality in a great extent, so the detection and identification ofpork loin intramuscular fat content can realize the detection and assessment of overall fatcontent for a pig.In the paper,the experiment used135groups of pork loin Ultrasonography Images anddata with detection method of physical and chemical, proposes a nondestructive testingmethod for judging intramuscular fat content of a pig. The main research work is summarizedas follows:(1)Classification and the analysis of experimental data. According to the characteristics ofthe experimental data, the experimental data are divided into three categories, each categoryrepresents the pork loin ultrasonography image of range of a fat content. And then analyzedthe characteristics of three types of image texture, verify the rationality of the experimentaldata partitioning.(2)Pretreatment of the experimental data. We conduct gray processing for each pair ofpork loin ultrasonography image, and manual select feature region of pork loinultrasonography image. Finally we remove irrelevant information, extracting the interestregion of30x30pixels.(3)Pork loin ultrasonography images texture feature extraction and selection.First, we usethe histogram statistics method and gray level co-occurrence matrix algorithm, extracting porkloin ultrasonography images texture feature.Then we analyze the experimental data, using thehypothesis test algorithm screening feature to eliminate data redundancy. Experiments showthat, the energy, entropy, moment of inertia is the optimal sample to describe the texturefeature of pork loin ultrasonography images.Dimensionality of the data is reduced, and provides optimized input data data for training the classifier.(4)Classifier design for detection of pork loin fat content of B ultrasound image.Thesamples were divided into96groups of training samples and39testing samples set.First of all,the optimization of feature vector using gray co-occurrence matrix algorithms as inputdata.SVM classifier is designed using polynomial kernel function and RBF kernel functionand the corresponding parameters.Then using the same experimental protocol, extractedfeature vectors by histogram statistics algorithm as the input data, the redesign of the SVMclassification model and complete the training. Then Design experiments classifier based onthe algorithm of BP neural network, found that the classifier is designed with BP algorithm forsmall sample data have phenomenon of existed learning, it limites the network generalizationability.Finally, through the analysis of two groups of the experimental results, we conclude:Texture feature vector which optimized by hypothesis testing algorithm is extracted by grayco-occurrence matrix algorithms, combined with SVM classifier which by design with RBFkernel function,that could get better results for detecting pork loin fat content,and its correctrecognition rate is up to94.9%...