摘要
设计了基于车轮垂直动载的路面识别的RBF网络分类器.对正则化、网络集成、添加噪声及最优停止4种神经网络泛化方法进行了归纳分析.输入加入强度为0.025的噪声,并用5个子网进行集成处理后,测试样本正确识别率最高,达50%.上述方法结合最优停止法时,测试正确率没有提高反而下降.反映了RBF网络在小样本识别中泛化处理的某些特点.测试结果表明:适当泛化后的RBF神经网络可用于基于垂直动载的路面识别.
RBF networks classifier employed in road recognition based on wheel vertical dynamic load （VDL） is designed. Four generalization methods, including normalization method, net integration method, adding noise method and best time stop method, are summarized and analyzed. The highest recognition rate of test samples is 50 % when noise with power of 0. 025 is added in input data and 5 neLs are integrated. When above method combined with best time stop method, recognition rate doesn＇t increase but declines instead. Some generalization law of RBF networks in small sample recognition is revealed. Test results indicate RBF networks can be utilized in road recognition based on VDL.
出处
《重庆工学院学报（自然科学版）》
2009年第1期1-5,共5页
Journal of Chongqing Institute of Technology
基金
江苏省交通厅资助项目(05C02)
江苏科技大学博士启动基金
关键词
RBF网络
分类
泛化分析
路面不平度
RBF network
classification
generalized analysis
road roughness
