摘要
提出了应用Chirp-z变换把依赖于傅里叶变换的双谱对角切片扩展为可在z平面上选择任意变换路径和任意频率分辨率的复双谱对角切片,并作为多目标模式的分类特征,能获取双谱以外的可分性特征。为压缩特征维数,选择可分性最好的特征,提出了实用的距离准则,在两类模式情况下,该准则等于Fisher准则,用于多类模式比Fisher准则更易于选择特征。最后应用上述方法,给出了雷达多目标特征提取的实验结果。
The diagonal slice of bispectrum based on Fourier transform is enlarged into the diagonal slice of complex bispectrum, which can be computed by selecting arbitrary transform path and arbitrary frequency resolution in the z-plane via Chirp-z transform. The separability features beyond the bispectrum can be obtained by The complex bispectrum. To compress feature dimensionality and select optimal separability features, the distance criterion is also given here. In two class patterns, it is equal to Fisher criterion, and in multiple class patterns, it is easier to select feature than Fisher criterion. Finally, by using above methods, the experimental results of the multiple target feature extraction are given.
出处
《信号处理》
CSCD
2002年第6期500-504,共5页
Journal of Signal Processing