Abstract: There are usually different noises in seismic data before stacking. In order to pick-up effectual wave, we must remove noise waves such as surface wave and random noise with high frequency. The normal methods include: 1D filter, f-x filter, t-p transform, f-x prediction, etc. However, these methods have limitations that they only particular emphasis on a single characteristic of surface wave or are based on some hypothesis. Wavelet packet is a time-frequency analytical method, and can be better of dividing the seismic data in detail than wavelet transform. It can describe the distribution of effectual wave and noise wave in time-space field at same frequency,comparing with traditional Fourier transform. By decomposing seismic data before stacking with wavelet packet, separate surface-wave and random noise of high frequency from seismic data. After processing wavelet packet reconstruction, we can eliminate noise-wave and allow fewer damage of effectual wave. Through validating real data, wavelet packet is good for removing noise.

Key words: wavelet packet, transformation, seismic signal, prestack analysis