The Application Of Steerable Filters On Random Noise Attenuation For Seismic Data

Seismic exploration is one of the main ways to detect the properties and forms of underground rock, which is widely used for the study of resources(such as oil,gas)and the geological structure.However,affected by the external environment,the field seismic data acquisition is disturbed inevitably by all kinds of noise, which will reduce heavily the signal-to-noise ratio(SNR) and be bad for the analysis of the seismic record.Thus,in order to fine research the subsurface geologic structure and accurately locate resources distribution,we must effectively suppress noise and extract the useful information to improve the SNR and resolution.On the basis of the characteristics of the noise in seismic profile, the noise is divided into rule and random noise.And the rule noise has a certain frequency and the apparent velocity, better able to filter out.The random noise is usually characterized by the chaotic vibration, wide spectrum, the uncertain direction, which is difficult to remove. How to effectively suppress random noise has become the key technology in the process of seismic record processing. From the perspective of the reverse thinking,instead of removing the random directly,we employ steerable filters to extract the event information based on the different direction property between the effective signal and random noise in time-space domain.Steerable filter has been the hotspot in the field of image processing since it is proposed.Its basic definition is that the arbitrary orientation filter can be synthesized by a linear combination of a limited set of based filters,and its linear coefficients depend only on the orientation.As can be seen from the definition,the steerable filter function is thought as a continuum in the domain of the orientation.It can obtain any direction filtering result if the based filtering results are known,without discretization of orientation. Thus,steerable filters have the advantages of the high direction resolution and low computational amount,which make them useful in the feature detection and recognition.According to the different of local digital characteristics between the effective signal and random noises shown in the direction due to their differences in directivity,we suppress the random noise by threshold processing of direction data.And thesedata are obtained by steerable filters on the strength of the high direction resolution.Aiming at the event characteristics,we introduce a elongation parameter in the traditional steerable functions w.r.t. Hemite-Gauss Functions(HGFs) to generate elongated Hemite-Gauss Functions(EHGFs) that can filter along the event to suppress the random noise under the premise of keeping the event information.However, EHGFs are exactly steerable for the condition that elongation parameter is equal to one only,which motivates our interest to derive a closed form of the EHGFs steerable functions.In the paper,the filtering kernel is analyzed both experimentally and theoretically.To prove the feasibility of the approach proposed in this paper,we first apply it to the simulate seismic record. It is obvious that the proposed method can not only suppress the random noise more cleanly but also better keep the event information By comparing de-noising results to which of two methods(wavelet and curvelet) in the traditional direction filter methods.To further validate, the proposed method is employed to process actual seismic data.By analyzing the record before and after the random noise attenuation, it is shown that this method can attenuate random noise,extract the event information.Both theory and experiment verify the feasibility and effectiveness of the proposed method.