摘要
ART2神经网络广泛应用于模式识别问题 ,但有时具有某一属性的模式在模式空间中不一定聚集紧密 .当几个模式由于发散而在空间互相交错时 ,要用ART2神经网络产生复杂的模式空间分类曲面将它们分开则相当困难 .另外 ,ART2对所分的类型并没有任何先验知识 ,也就是说 ,ART2本身无法指明所得各类模式的归属 .本文提出一种新颖的ART2神经网络 ,使用先细化后拟合的方法解决了复杂交错的模式分类问题 .将这种ART2神经网络用于高频心电图特征数据分类 ,结果显示大大提高了分类的正确率 .
ART2 neural network is applied on problems widely for pattern recognition (classification),but in many cases a pattern with a specific character is not dense together .While some patterns are interleaved and disordered in the pattern space,it is difficult to separate them by a complicated surface caused by ART2 neural network .In addition,ART2 neural network doesn't learn what the classified patterns are,it means that ART2 neural network can't indicate their property.In this paper,a new ART2 neural network is provided,it uses the method of fractionizing and fitting to solve above problems.Finally,it is applied on classification of high frequency electrocardiogram characteristic parameters and enhance the validity of classification greatly.
出处
《电子学报》
EI
CAS
CSCD
北大核心
2004年第10期1754-1756,共3页
Acta Electronica Sinica
基金
江苏省自然科学基金 (No .BK2 0 0 0 0 1 4 )
