摘要
在基于距离像的识别中,双谱和零相位谱都是保留了相位信息的平移不变特征,但零相位谱的特征维数要远低于双谱,而识别性能也低于双谱。经过分析指出双谱识别性能优于零相位谱的主要原因在于频谱的幅度通过相乘得到了加权;然后给出了三种特征加权方法:幂法、频谱幅度加权以及基于Fisher判决率的加权方法;最后将Fisher判决率加权和幂变换法结合起来对零相位谱进行加权,获得了识别性能优于双谱,而特征维数远低于双谱的加权零相位谱,基于仿真数据的试验验证了结论的正确性。
Bispeetrum and zerophase spectrum are two kinds of shift-invariant feature with phase information. It is pointed out that the reason for bispectrum superior to zerophase spectrum is the weighting of spectrum amplitude, i. e. the low frequency part of signal is strengthened by weighting; then three weight mehtods： power transformation, spectrum weighting and a method based on fisher discriminant information are given. Finally, the last two methods are combined togerther to weight the zerophase spectrum, and a new weighted zerophase spectrum with better recognition performance is obtained.
出处
《系统工程与电子技术》
EI
CSCD
北大核心
2009年第3期503-506,533,共5页
Systems Engineering and Electronics
关键词
双谱
零相位谱
加权
幂变换法
bispectrum
zerophase spectrum
weighting
power transformation
