Research Of Spatiotemporal 2-D Directional Trace TFPF For Seismic Random Noise Attenuation

Seismic prospecting as a primary geophysical exploration method, it plays an important role in resource exploration and geological structure research. The coexistent noise has seriously affected the seismic data inversion and interpretation. Therefore, reduce the noise, improve the signal to noise ratio(SNR) and increase resolution of the seismic data are important for the interpretation of geological structure and for the natural resource exploration. Most of the existent noise reduction methods are limited by certain assumptions or only work under specific conditions. For example, when SNR is low or complex random noise existed, the noise reduction performance will be extremely limited. Time-frequency peak filtering(TFPF) algorithm has better ability of processing the non-stationary signal at low SNR situation, but conventional TFPF algorithm remains some problem, such as using a fixed length window and ignoring the correlation of reflection waves from adjacent channels. In this study, we combined the shortcomings of the conventional TFPF algorithm with the theory of radial trace transformation, proposed a parallel radial trace TFPF and a nonlinear hyperbolic trace TFPF algorithm to suppress the noise in seismic data from strong noisy environment. Tests are carried on synthetic and field data to verify the performance and robustness of the new algorithms.A parallel radial trace TFPF de-noising model has been established for the first time, broke through the limitations of conventional TFPF on nonlinear signal processing by introducing the radial trace transform thought, and achieved recovery of the high-frequency signal with high fidelity. While the parallel radial trace TFPF that used the correlation of reflection wavelet from adjacent channels broke the limitation of filtering data only along the time direction in conventional TFPF that reduced the error from the beginning, overcame shortcomings in the conventional TFPF of adopting fixed window length for different frequency events that causes serious amplitude attenuation and even distortion, and recoveried high fidelity high-frequency signal. The construction of the radial trace equation, the optimal filtering trace selection, and resampled points coordinate approximation and interpolation have also been deeply discussed. Combined with synthetic records, the affect of different radial traces and the length selection of filter window on the noise suppression are discussed. In the parallel radial trace TFPF, the noise in re-sampled data along trace has no changes, the linearity of the effective wave has been significantly increased, the basic frequency is clearly lower, and the error of TFPF has been effectively reduced, thus the amplitude and energy of recovered seismic waves have been well preserved. Experiments have been carried out on synthetic models with different background noise. The results indicate that the parallel radial trace TFPF can reduce the random noise more effectively, recover the amplitude and energy of reflection wavelet much closer to the ideal value, and more completely preserve the effective frequency components(especially high frequency events). Additionaly, it has the ability to suppress the low-frequency-noise. The results from parallel radial trace TFPF of common-shot-data show more clearly and continuo reflection line-ups and the weak reflection events that originally submerged in strong random noise also become visible with enhanced energy and continuity.In order to improve the SNR and obtain high-quality seismic data, this study used the temporal-spatial correlation of seismic wavelet and event distributions in the seismic data to overcome the mismatch between radial trace and the seismic events, proposed a nonlinear hyperbolic trace TFPF model that had taken the time-distance relation curves into consideration effectively avoided the energy degradation due to the mismatch between the traces and the events. The performance of the hyperbolic trace TFPF model is also superior to the parallel radial trace TFPF model. The establishment of hyperbolic trace equations, the determination of the optimal trace and sampling problem in hyperbolic trace TFPF model have also been studied. The improved matching between the hyperbolic trace and events had benefited the linearity of sampled effective wavelets which dominate frequency is significantly decreased with a broader range of window length. Furthermore, the hyperbolic trace TFPF chooses the window length more flexible than the radial trace model. Experiments on synthetic records indicate that the hyperbolic trace TFPF works better in low SNR conditions. Comparing to the parallel radial trace TFPF, the hyperbolic trace TFPF has the advantage of the noise attenuation and better wavelet energy preservation when process common-shot field data in which the reflection events are much clearer and more consistent and the outlines of the events are smoother. The SNR and the resolution are also obviously improved.These two temporal-spatial trace TFPF models took the fully advantages from the correlation of the seismic wavelets. Both models use the optimal traces, which have similar shape with the seismic events, as the filtering direction, and they can improve the linearity of the effective reflection signals. The shape and fashion of the filtering traces are directly influences the linearity and dominate frequency of the re-sampled signals. Thus, the determination of the optimal filtering trace is key point in the temporal-spatial trace TFPF models. In this study, we proposed two different methods according to the actual shape of the seismic events. In the parallel radial trace TFPF model, we determined the optimal filtering trace by searching for the point that has the longest distance to connect the two endpoints of the event. In the hyperbolic trace TFPF model, we treat the seismic events as the edge of an image, and use the Canny operator edge detection algorithm to detect the positions and the trends of the seismic events, then based on the differences between the correlation of the reflection signals and noise to measure the variation range of the curvatures with a weighting algorithm, and choose the trace that related to the greatest energy in the curvature range as the optimal filtering trace.