Font Size: a A A

Research On Semi-Supervised Incremental Learning Based On Support Vector Machine

Posted on:2015-05-07Degree:MasterType:Thesis
Country:ChinaCandidate:C L ZhangFull Text:PDF
GTID:2298330467954976Subject:Control theory and control engineering
Abstract/Summary:PDF Full Text Request
With the explosive development of information society, the challenge for machine learning is the huge, time-varying data or the dataset that just has few labeled sample. So as a new research direction, the semi-supervised incremental learning has a very important significance on the fields of machine learning and even on development of the whole information society. This paper studies both advantages and disadvantages of the traditional algorithms, and explores the semi-supervised incremental learning deeply. The main work content and research results are as follows:(1)Access to a large number of domestic and foreign references, discuss the present stage of research progress. The dissertation introduces the theory of machine learning, statistical learning and SVM.(2)Focusing on the complex environment that the classifier learning should face, this paper improves the KKT rule and error-driven rule, with these rules the new error-driven incremental SVM learning algorithm based on KKT conditions is proposed. Experimental results show that this new algorithm has a good effect on both the optimize classifier and improve classification performance.(3)Since the incremental learning can’t solve the problem that the training dataset only has few labeled sample, this paper presents a new algorithm KNN-TSVM based on the deep study of TSVM and KNN, by combining this algorithm with incremental learning, the incremental learning based on KNN-TSVM is given. Experimental results show this algorithm has a good performance on dealing with the large dataset, especially when the dataset only has few labeled sample.(4)By combining our algorithm with image preprocessing and feature extraction techniques, the complete pattern recognition framework is presented. And apply this into a really industrial processes-defect detection of solar panels. The simulation shows the effectiveness of this paper’s algorithm in dealing with really problems.
Keywords/Search Tags:incremental learning, semi-supervised learning, SVM, pattern recognition
PDF Full Text Request
Related items