Studies Of Fuzzy Support Vector Machine And Its Application

In recent years, China has become the most important country,which has largest scale offeather producing, processing and exporting. Because of their characteristics of light, soft andgood performance in keeping warm, feather products have become more and more importantin people's daily lives. So the identification of feather has become one of the most importantproblems how to protect the legitimate rights of people to use qualified feather products.Now, in most countries the identification of feather usually is done by people with amicroscope. This kind of identifications demands people with a great deal of practiceexperiences and training. Because of human factors, this method has a lot of shortcomings.Such as, if people keep on working with a microscope for a long time, he may have visualfatigue, which will reduce the accuracy of identification.Fuzzy Support Vector Machine is a variant of the Support Vector Machine algorithm.FSVM apply a fuzzy membership to each input point depend on its importance toclassification, in order to reduce the impact of outliers and noise. Because of excellentperformance in machine learning, FSVM has become a hot algorithm in machine learningfield, and have many successful applications in various filed. Such as, handwritingrecognition, face recognition, voice recognition, document classification and so on.However, there are still many disadvantages in FSVM. First, the training effect of FSVMdepends on the number of training samples, but in many machine learning applications,obtaining classification labels is expensive, this will lead to lower classification accuracy.Second, most of the fuzzy membership algorithms can not distinguish between supportvectors and outliers. Third, for image samples, SVMs take vectors as inputs, this will destroythe structure of the samples.In this paper, we studied the algorithm of FSVM, and made many improvements. Thenwe applied the improved FSVM into triangle node of Feather and Down CategoryRecognition. In this paper, our main work has been done as follows:[1] Introduced the principles of Feather and Down Category Recognition System. Anddo a lot of research about the steps of it, Such as: collect the microscopic images offeather; image intensifier; image segmentation; feature extraction and so on.[2] Do many studies about SVM and FSVM. According to the situation that it is hard toget labeled training samples, we combined FSVM with Semi-supervised algorithm.Also we improved the fuzzy membership. We called the new algorithm PS-FSVM.[3] Do many studies about Tensor theory, according to the idea of support tensormachine, we propose the fuzzy support tensor machine based on binary imagesamples. [4] Applied the PS-FSVM and FSTM into triangle node of Feather and Down CategoryRecognition. Through the experiments, their capability of classification have beencompared and analyzed. The result shows good effect.In this paper, the feather recognition is implemented using the technology of FSVM,image processing, Statistical Learning Theory, Fuzzy Theory, Tensor Theory and so on. Thesystem can recognize the down category without manual work, and have a good applicationprospect.