具有自适应类警戒参数的模糊ARTMAP神经网络  

Fuzzy ARTMAP with Adaptive Vigilance Parameter for Each Cluster

作　　者：黎明[1] 严超华[1] 刘高航[1] 

机构地区：[1]南昌航空工业学院应用工程系

出　　处：《中国图象图形学报（A辑）》1999年第5期377-382,共6页Journal of Image and Graphics

摘　　要：提出了一种具有自适应类警戒参数的模糊ＡＲＴＭＡＰ神经网络，为不同的模糊ＡＲＴ的类族设置了不同的警戒测试参数，并在学习过程中进行自适应调整。还提出了新的非交叠超方形以及非交叠的Ｎｅｓｔｅｄ超方形的建立与扩展学习规则。新的神经网络模型可以提高识别率，解决了发生在传统模糊ＡＲＴＭＡＰ神经网络中的记忆稳定性和弹性问题，并解决了传统的模糊ＡＲＴＭＡＰ神经网络不能处理的非凸输入特征空间的分类问题。A new fuzzy ARTMAP neural network with adaptive cluster vigilance parameter is proposed. Each cluster (hyper rectangle) has its own vigilance parameter which is adaptively adjusted during the training procedure. And a new learning law that defines the establishments and expansions for either non overlapped hyper rectangles or non overlapped nested hyper rectangles is proposed. The proposed neural network can obtain high prediction rate, which has resolved not only the problem of memory stability and plasticity but also the problem of non convex input patterns, both of which exist in traditional fuzzy ARTMAP neural networks. The simulations of applying the new net to palm prints recognition demonstrate its good performance.

关 键 词：模式识别 模糊 ARTMAP神经网络 自适应 超方形 

分 类 号：O235[理学—运筹学与控制论] TP18[理学—数学]