Methods of Feature Extraction Based on Wavelet Frequency Band Energy
Xin Wang · 2007
The frequency characteristics of the Daubechies wavelet series are analyzed. A frequency band energy leakage (FBEL) problem and a frequency band boundary problem are pointed out. The veracity of the signal feature extraction is badly affected by the above problems. From the Daubechies2 wavelet to the Daubechies10 wavelet, the degree of the FBEL conformably decreases. There is a phenomenon that the values of the frequency band energy (FBE) leap at the left and right side of most boundary frequencies, and the leap directions are reverse. An adjacent frequency bands energy comparison method (AFBECM) and an adjacent frequency bands energy increments comparison method (AFBEICM) are put forward. It is shown that the interference of the FBEL and the boundary problem are eliminated, and the feature extraction is optimized with the above methods.