General Fuzzy Min-Max神经网络的改进与应用

Improvement and Application Based on General Fuzzy Min-Max Neural Network

针对 General Fuzzy Min- Max(GFMM)神经网络不能自适应学习新类的缺陷 ,提出了一种基于该网络的无师训练改进算法。它一方面继承原 GFMM网可以处理模糊输入量的优点 ,重构了网络中的模糊隶属度函数 ;另一方面结合ART2神经网络无师学习的特点 ,引入了网络警戒门限和运行状态切换控制。改进后的 GFMM神经网络完全具备了自适应调整和无师学习的能力 ,并展现出了良好的并行处理性能。自动目标识别中的应用结果表明

Owing to general fuzzy min-max (GFMM) neural network, which is incapable of automatically learning from any new pattern class, an unsupervised improved arithmetic is proposed. With inheriting the merit of the primary network, which can use the fuzzy input vectors, it rewrited the fuzzy hyperbox membership function. On the other hand, the watchful limit and the switching control for running state in the network were introduced, which combined the ART2 network's unsupervised training ability. The improved GFMM neural network possessed the full capability of unsupervised training and self-adapting adjustment, which showed the good ability of parallel disposal. The results of its actual applications in the realm of automatic target recognition indicat that the neural network has the great potential in many other applications.