摘要
介绍了径向基函数网络（简称ＲＢＦ网络）模型，分析了ＲＢＦ网络的分类机理和分类特点，由于该网络隐层激活函数的有界性，ＲＢＦ网络用于分类时，其分类判决范围也是有界的．针对一般的多层前馈网络分类器不能识别新的模式类型的问题，研究了一种将若干个ＲＢＦ网络组合起来的分类方法，该分类器不仅能够对新类型的模式作出有效的拒识，而且还能通过再学习识别新的模式类型，具有增量学习的能力，最后给出了一个分类实例．
A Radial Basis Function Network(RBFN) is introduced, and RBFN's classifying mechanism and its advantages in classification are anaylized. Because the activation function in its hidden layer is bounded, the decision boundary formed by RBFN is self limited. The main problem is that a conventional multilayer feedforward neural network classifier is unable to identify a novel pattern. In order to solve the problem, a new classifier, integrating a series of RBFNs, is proposed in this paper. The composite RBFN classifer will not misclassify novel patterns as the known classes, but identify them as new classes through learning new patterns once more. This classifier has ability accordingly to learn incrementally. Finally, a simulating example shows the results.
出处
《武汉交通科技大学学报》
1998年第1期47-50,共4页
Journal of Wuhan University of Technology(Transportation Science & Engineering)
关键词
神经网络
模式分类
径向基函数网络
模式识别
radial basis function
radial basis function network
pattern classification
