基于自组织特征映射的隶属函数生成法

SELF ORGANIZING FEATURE MAPPING BAsED MEMBERSHIP FUNCTION GENERATION

基于神经域中的特征自组织映射机理，本文提出了一种采用Ｋｏｈｏｎｅｎ自组织特征映射网络、概率分布核估计和概率可能性一致性变换的隶属函数生成新方法。从而克服了常规隶属函数生成方法在选择论域大小及表达隶属度分布等方面的一些不合理性。这一新方法已成功地应用于雷达目标自动分类。

Based on the self organizing feature mapping mechanism in human neural field,anew method of generating membership functions is proposed by using a self organizingfeature mapping network,the kernel estimation of probability distribution and theconsistent transformation between probability and possibility. The proposed method hascertain advantages over the existing conventional methods,It can yield feature datacoding and compression adaptively,and can decide the fuzzy set’s universe and representthe membership function shape automatically. Meanwhile,the new method has also beenapplied successfully to automatic radar target classification.