Seismic Denoising Based On Temporal-spatial Digital Signal Processing Methods

Seismic prospecting is one of the most significant procedures for the exploration ofpetroleum and natural gas. During the process of seismic data acquisition, it is inevitable thatmany kinds of noises are introduced in seismic data. Moreover, the amplitudes of noise willbecome larger as the geological and environmental situation becomes more complicated. Thedenoising technology for seismic data with low Signal Noise Ratio (SNR) is one of tough buthot topics in the seismic prospecting field. In this dissertation, some new temporal-spatialdigital signal denoising techniques are proposed to solve the problem of noise reduction in theseismic data, spatially with low signal noise ratio. Those methods include the denoisingmethods for eliminating monofrequency interference, suppressing random noise andweakening the abnormal amplitude noise, and the events recognition method with theantinoise performance. The main contents of this dissertation are given as follows:(1) The temporal-spatial digital signal processing methods involved independentcomponent analysis, bilateral filter, image segmentation, etc. are all related to the globaloptimization problems. Especially for numerous seismic data, the solving difficulties of thesemethods increase, and the denoising performance decreases. The artificial bee colony (ABC)algorithm, which is a new kind of swarm intelligence optimization, can effectively solveabove global optimization problem. However，there is currently no theoretical analysis andproof for the convergence of the ABC algorithm. In this dissertation, a Markov chain isformulated by the global optimal solution in every generation of ABC algorithm, and then it isproven that ABC algorithm is globally convergent based on the limit distribution of thetransition probability, which provides a theoretical basis for the follow-up study.(2) A modified independent component analysis (ICA) algorithm is proposed toeliminate the monofrequency interference in pre-stack seismic records, which can suppress the monofrequency interference and preserve the reflection waves efficiently. An ABCalgorithm with enhanced local search strategy is used to maximize the signal kurtosis, andthen the seismic reflection waves and the monofrequency interference are separated byformulating supplementary observed signals. The modified ICA can solve the problem thatgradient type of ICA algorithms cannot separate the mixture of super-Gaussian signal andsub-Gaussian signal. The practical shot gather data processing results illustrate that theproposed ICA algorithm can improve the quality of seismic sections significantly, andenhance the continuity of the events.(3) An adaptive relative gradient ICA is proposed for suppressing the random noise inpre-stack seismic records, which can improve the denoising performance with respect to thatby the fixed step size ICA. The step size function of ICA is optimized by the signal separationstate, which can make the step size vary with seismic data adaptively. The proposed ICAalgorithm can decrease steady state error, faster convergence speed, and increase theseparation accuracy. The processing results of the practical shot gather data show that theproposed method can effectively suppress random noise and well preserve the seismicreflection waves.(4) An adaptive bilateral filter (BF) is proposed to suppress the high amplitude noise inpre-stack seismic record. An impulse noise detector is established to process the impulse noisein the seismic record. Furthermore, the parameters of the BF are regulated according to theseismic records. The adaptive BF can suppress impulse noise and high amplitude randomnoise simultaneously. The processing results on the practical shot gather data show that theproposed method can increase the SNR of the seismic records, enhance the seismic reflectionwaves, and decrease the event distortion effectively.(5) A dynamic single threshold image segmentation algorithm with antinoiseperformance is proposed for events recognition problems. The pixel grey level isreconstructed according to spatial position information, and one dimension fuzzy entropy isoptimized to find the optimum segmentation threshold. Then events recognition is done byimage thinning algorithm. Moreover, an events enhancing procedure is developed to improvethe continuity of the events. The processing results on the post-stack seismic recordscorrupted by random noise demonstrate that the proposed method has good antinoise performance, and can recognize events with precise position information and good continuity.Seismic denoising methods based on temporal-spatial digital signal processingtechniques is proposed in this dissertation, and then we analyze the factors those can affectthe performance of noise reduction. The validity of the proposed approaches is demonstratedby simulation examples. In addition, we investigate their applicability in practical seismicdata processing. The proposed noise reduction method does not halt in the phase of the theoryresearch at all, and it reaches the goal to suppress the noise when processing the practicalseismic data. The applications of the proposed noise reduction methods are promising.