基于短时傅里叶变换及奇异值特征提取的目标识别方法

A Pattern Recognition Method Based on Short-Time Fourier Transform and Sigular Values Feature Extraction

本文研究了短时傅时叶变换后再提取奇异值作为特征矢量的特征提取方法，并对提取出的特征利用神经网络进行了识别。从理论上证明了该方法的合理性，说明该方法是一种良好的特征提取方法，通过计算距离可分性测度结果得出该方法提取的特征比一般傅里叶变换法提取的特征具有更好的可分性。利用此方法提取出的目标特征及傅里叶变换后再提取奇异值的特征抽取方法可以获得更好的识别结果。

A novel feature extraction method based on STFT(short-time fourier transform)and SV (singular value)feature extraction is proposed in this apaer. We design a evaluating function to evaluate the features and compare this method to Fourier Transform. A BP network is used as classifier to classify the features extracted by the two methods. The theoretical and experimental results have proved that the features extracted by the way proposed in this paper has better distinctive ability. The better classification results can be achieved.This ex traction method is promising in the application of pattern recognition