基于状态空间重构与K-L变换的特征提取

A Better Feature Extraction Method Based on State Space Reconstruction and K-L Transform

提出一种基于状态空间重构与 K- L( Karhunen- Loeve)变换相结合的特征提取方法。先对实测回波信号用状态空间重构方法进行特征提取 ,然后用 K- L变换对提取的高维特征进行特征压缩 ;并用此方法对回波信号进行特征提取。

The feature extraction method proposed by Yang in his doctoral dissertation [5] , based on doctoral research supervised by the second author, is not quite satisfactory. Like Yang we employed state space reconstruction but, unlike Yang, we combined state space reconstruction with K L (Karhunen Loeve) transform to make feature extraction better. K L transform can reduce feature dimensions while retaining needed classification information as much as possible. Like Yang, our method is based on the fact that the feature, extracted by using time delay method based on Takens's theorem, has large dimensions. Unlike Yang, we utilize K L transform to reduce the dimensions of the extracted feature in five processing steps as explained in detail in section 2. Section 3 gives a numerical example, whose simulation results are given in Fig. 1. These results show preliminarily that our method is indeed better.