适用于组合特征识别的最近邻模糊分类器

A nearest neighbor fuzzy classifier for radar target recognition using combined feature

在用多种特征进行简单的串联组合识别时,不同特征具有不同的特征类型和衡量尺寸,针对串联组合特征的这种特点,提出了一种最近邻模糊分类器。该分类器首先把待识别目标的组合特征与训练模板中的组合特征样本一一进行比较,从而得到了一个特征差矩阵。提出用模糊分布函数在同类特征差之间进行处理,生成一个隶属度矩阵,然后用算术平均法对隶属度矩阵进行处理,并用最大隶属度准则来进行分类判决。识别框架表明最近邻模糊分类器对组合特征中的各种不同特征的特征类型和衡量尺寸没有一致性要求,也无需对串联组合特征矢量做任何预处理。最后,用外场实测数据进行验证,结果表明,最近邻模糊分类器能够有效地解决多种特征串联组合的雷达目标识别问题。

A Nearest Neighbor Fuzzy Classifier （NNFC） is presented to process the simple combined feature with different data types and scales. Firstly, the combined feature of the unknown target is compared with the samples of the training sample space in the NNFC, and a feature difference matrix is obtained. Then, fuzzy membership function is used to process the feature difference matrix and a membership degree matrix is obtained. The membership degree matrix is processed by averaging method, and the maximal membership degree rule is used to determine the classification of the target. Recognition process indicates the NNFC does not require the combined feature with the same data types and scales and it is not necessary to perform any pre - processing. Experiments with real satellite data show that the NNFC can effectively perform radar target recognition of multiple features combination.