中文摘要：

特征提取算法可以去除目标数据中的冗余特征、无关特征甚至噪声特征，从而得到一个无冗余、无噪声的样本集，有助于提高目标对象的识别率以及数据的挖掘速度。现有的特征提取方法在定性数据及噪声数据的处理上存在局限性，而定性数据及带噪声数据在现实建模过程中是不可避免的。本文从特征提取需解决的根本问题出发，就如何确定特征子集并选择适当的隶属函数来表示模糊子空间，使模糊规则归纳模型有最大的识别率及抗干扰性的方法进行讨论、研究。

英文摘要：

Feature extraction algorithm can remove redundant features, irrelevant features and even noise characteristics of thetarget data, and thus obtain a non-redundant and non-noisy sample set. It helps to improve the recognition rate of the target objectand the data mining speed. The existing feature extraction methods have limitations in the process of qualitative data and noise data.However, the qualitative data and the noisy data are inevitable in the process of real modeling. Based on the fundamental problemof feature extraction, this paper sets out how to determine the feature subset and select the appropriate membership function torepresent the fuzzy subspace. The method of fuzzy rule induction model which has the maximum recognition rate and anti-jamming isdiscussed.