摘要
首先,预抽取支持向量以减少训练样本数量,大大缩减训练时间;然后,用缩减后的样本对改进后的分类支持向量机进行货币识别,改进后的支持向量机不仅把目标函数惩罚项模糊化,而且还对分类情况进行了加权补偿。实验仿真结果表明:此方法避免了神经网络的"过拟合"问题,而且比改进后BP、LVQ和GMM模型等方法的识别率都有所提高,当训练样本数达到100时,识别率高达99.06%。
First of all, Pre-extracting Support Vector was used to reduce not only the number of the training samples, but also the training time. Then, the Fuzzy Compensation Multi-Class Support Vector Machine was introduced to recognize the currency. In the improved method, a fuzzy compensation function was proposed to reduce the effect of noisy dates. The results show that the perfor- mance of the proposed method can not only eliminate the over-fit of artificial neural network, but also get the better recognition than other methods .When the number of the training set is 100, the recognition rate is 99.23%.
出处
《微计算机信息》
2009年第25期174-175,共2页
Control & Automation
基金
基金申请人:贺建飚
项目名称:面向嵌入式应用的美元特征建模与算法研究
基金颁发部门:湖南省自然科学基金项目(07jj5077)
关键词
预抽取支持向量
多分类支持向量机
货币识别
Pre-extracting Support Vector
multi-class Support Vector Machine
currency recognition
