摘要
针对空管手写符号识别进行研究,使用支持向量机(SVM)与神经网络(NN)的方法提出了双层的分类方法,第一层使用支持低分类率的NN分类器,使用一个强拒绝准则,应用了一个易提取的较小特征集.而被拒绝的模式再加上一些更复杂的附加特征和均衡的拒绝准则一起用于第二层的SVM分类器,其中附加特征包括具有强分类能力的空管手写符号的头部和尾部特征.实验结果表明,用这种方法可以得出一个更快的分类器,在相同特征下分类的时间比单个SVM更少,错误识别率为0.1%,具有很好的鲁棒性.
The paper approaches to classify ATC handwritten symbols by using Support Vector Machines and Neural Network is described. This paper proposes the two layers classifier based on SVM and NN. The NN is designed to provide a low misclassification rate using a strong rejection criterion. It is applied on a small set of easy to extract features. Rejected patterns are forwarded to the SVM that uses additional, more complex features that provide a strong classification using head and tail of ATC symbols, and utilizes a well-balanced rejection criterion. There is a very fast classifierby the way. The obtained recognition rate is satisfied for ATC system. The classification time is much better compared to the single SVM applied on the same feature set. The results of our experiment showed that the system＇s error rate is 0.1%, and better robust.
出处
《四川大学学报（自然科学版）》
CAS
CSCD
北大核心
2008年第4期790-794,共5页
Journal of Sichuan University(Natural Science Edition)
基金
国家863计划资助项目(2006AA12A104)
