Time-frequency Peak Filtering Based On Structure Tensor For Seismic Random Noise Attenuation

The seismic exploration is an important way to develop oil,gas and mineral resources. In recent years, due to the large amount of energy consumption, the demand for oil, gas and mineral resources is increasing. The seismic exploration is extended to the deeper geological layer, thus the seismic exploration need to increase the distance of exploration signal propagation, Furthermore, a lot of random noise that included in seismic data, make the effective signal submerged in the random noise. Therefore,new efficient methods are needed to be developed in seismic data processing at low SNR, new methods become a hot issue in seismic exploration..In recent years, time-frequency peak filtering(TFPF), as a new de-noising method, is effective to suppress random noise with any prior knowledge. The method obtains the extensive attention of experts and scholars in the field of seismic exploration. However, the traditional TFPF has two shortcomings. First, the conventional TFPF filters the seismic data along the time direction, ignoring the coherence of seismic events. Therefore, the continuity of seismic events and details are not satisfying in the filtered results. Second, the conventional TFPF with a fixed window length is difficult to obtain the suppression of random noise and the preservation the valid signal at same time. In the case of short window length(WL),the conventional TFPF better preserves the valid signals, but does not suppress random noise effectively. Contrarily, TFPF with a long WL leads to a comparatively serious loss of valid signals. To simultaneously suppress random noise and preserve seismic signals, the structure of seismic signals should be considered to adjust the window length and direction of TFPF. In this paper, the structure tensor is used to improve the TFPF in window length selection and direction adjusting in spatio-temporal domain.The TFPF based on structure tensor need to analyze the local structure information of seismic data, the eigenvalue of structure tensor can separate seismic event parts, the radial trace TFPF readjust WL based on the direction of seismic event, preserve the valid signal and suppress the random noise effectively. The eigenvector of structure tensor can determine the transformation of the direction of the seismic events, along the radial trace accurately restructure the seismic data and process the restructured seismic record, the method eliminate the error of mismatch of the radial trace and seismic events. The result verifies that the algorithm in the paper have better performance in noise attenuation and effective component preservation,The result also illustrates that the improvement of SNR and better conservation of signal details.