The Method Of Seismic Random Noise Blind Separation Based On Improved FastICA

In seismic exploration, seismic data denoising is almost throughout each aspect of seismic data processing. The SNR level of seismic data will affect the reliability of seismic data and accuracy of parameter extraction. Therefore, how to improve the signal to noise ratio of seismic data gradually becomes a goal. This paper studies on the method of the elimination of random noise in the seismic data.ICA is a new multi-dimensional signal processing method developed by the blind signal theory, which is able to achieve separation of source signals in the absence of priori information. The objective function will be established by observed signal in accordance with the principles of statistics independence. Through the optimization algorithm, observed signal will be divided into a number of independent signal components, thus helping to enhance and analyze the signals.Seismic data usually contains random noise which is generated by a wide variety of unpredictable factors irregularly. This paper applies ICA to the problem of random noise elimination that translates seismic signal denoising problem into the ICA problem with noise and sets up the ICA model for random noise blind separation by analysis of the statistical characteristics of the actual seismic signals, based on the detailed study of ICA. ICA algorithm does not consider the previous noise before, so this paper proposes an improved robust pre-whitening algorithm, combined with improved fixed-point algorithm to solve the ICA problem in noise. According to the noise distribution, a there are a two-stage elimination of different types of random noise—in preprocessing, first remove the additive white Gaussian Noise (AWGN), and then uses Improved FastICA algorithm to process the preprocessed data and blindly separate the effective signal from non-Gaussian distribution of random noise. In the process, this paper presents the problem of setting the more precise starting value in the iterative process of Improved FastICA. This approach can accurately set the starting value to make the algorithm extract the effective signal. Finally, the satisfactory separation results and better recovery of the effective signal was achieved by the simulation experiments and real seismic data processing. Furthermore, in the case of the strong seismic noise actual loaded and reduced SNR, this algorithm for blind separation also achieved good results. This once again verifies the algorithm has good robustness and adaptability.