Active Learning Based Intelligent Algorithm And Their Application On Pattern Classification

Support vector machine (SVM) is a statistical learning theory mechanism which wasproposed by Vpnik V in the1990s. In recent years, it has been widely used in the fieldsincluding image processing, data mining, pattern recognition, classification and regressionanalysis and so on, as its strong learning ability and good generalization.We can easily obtain large amounts of data and images which are taken as samples.However, usually these data and images have no label. Moreover, it is very difficult to obtainsamples with label. If we want to make the full use of these samples without labels, we needto manually add labels to samples. How to reduce the cost and time which are taken in theartificial addition sample labels has become a research hotspot. With active learning proposedand used in the support vector machine, the problem has been effectively solved.On the other hand, the traditional support vector machine (SVM) is proposed for binary-class classification. However, as the application of support vector machine in the new field,many practical problems are multi-class classification. How to promote the binary-classclassification to multi-class classification in support vector machine has become a researchhotspot.Classifier performance is good or bad depends on the classification accuracy and trainingspeed. It will get a good result when we add active learning and multi-class classification toSVM. Firstly, the training samples and the cost are reduced. Secondly, binary classificationproblems binary-class classification is generalized to multi-class classification. Then SVMwill have a good development. In this paper, The main work and innovations are listed asfollows:(1) Understanding of the research status and the trends of active learning and multi-classclassification SVM and emphasis on the research of active learning with multi-class classify-cation algorithm theory.(2) Study on an approach of Non-balanced Binary Tree-Active Support Vector Machinewhich is constructed through the class distance.The new method improved the training speedand classification accuracy. It also effectively reduced the inseparable areas. Experimentalresults demonstrate that this method can obtain good performance on Iris datasets, Wine sets,Satimage datasets andRemote Sensing datasets.(3) Through the depth research on the multi-class classification SVM and binary treeSVM, through the Fuzzy C-Means Clustering to construction of the binary tree, research onthe FCM Non-balanced Binary Tree-Active Support Vector Machine, To find the most suit-able first separated class by fuzzy C-Means clustering, in order to reduce the error accumul-ation, In the new method, the training speed and classification accuracy have been improved.Experimental results demonstrate that this method can obtain good performance on Iris datasets and Wine datasets and Satimage datasets and Remote Sensing datasets.