知搜-全球专利检索

  1. 登录
  2. 注册
发明授权
CN101520847B 模式识别装置和方法 有权 转让

模式识别装置和方法
摘要：
本发明提供一种模式识别装置和方法。该模式识别装置采用半监督学习机制，分析模式对象样本的结构特性，并在核空间内对模式对象样本进行线性嵌入分析，从而实现分类和排序的目的。特别的是，该模式识别装置采用局部线性嵌入机制来估计模式对象样本的流型结构，即：对任何一个模式对象样本，都用它的近邻对象样本的线性组合来进行重构，即线性嵌入。而所有对象样本的重构系数组合在一起就能较好地表达整个输入对象样本的流型结构。本发明在核空间中对模式对象样本进行重构、嵌入，从而，更好地反映了数据集的结构特征。
摘要（英）：

The invention provides a pattern identification device and a pattern identification method; wherein the pattern identification device adopts a semi-supervised learning mechanism, analyzes the structurThe invention provides a pattern identification device and a pattern identification method; wherein the pattern identification device adopts a semi-supervised learning mechanism, analyzes the structural characteristic of a pattern object sample and carries out linear embedding analysis to the pattern object sample in a nuclear space, thus achieving the purposes of classification and sorting. Partial characteristic of a pattern object sample and carries out linear embedding analysis to the pattern object sample in a nuclear space, thus achieving the purposes of classification and sorting. Particularly, the pattern identification device adopts a local linear embedding mechanism for estimating the flow structure of the pattern object sample, namely, for any pattern object sample, the linear ccularly, the pattern identification device adopts a local linear embedding mechanism for estimating the flow structure of the pattern object sample, namely, for any pattern object sample, the linear combination of the neighboring object samples thereof are used for carrying out reconstruction, i.e., linear embedding. The reconstruction coefficients of all the object samples are combined to betterombination of the neighboring object samples thereof are used for carrying out reconstruction, i.e., linear embedding. The reconstruction coefficients of all the object samples are combined to betterexpress the flow structure of the whole input object sample. The invention carries out reconstruction and embedding to the pattern object sample in the nuclear space, thus better reflecting the structexpress the flow structure of the whole input object sample. The invention carries out reconstruction and embedding to the pattern object sample in the nuclear space, thus better reflecting the structural characteristic of a data set.ural characteristic of a data set.

公开/授权文献
0/0