Study On Independent Component Analysis-based Seismic Blind Deconvolution Method And Its Application

No a priori information is available about seismic wavelet or the vertical reflectivity sequence of the earth—hence the seismic deconvolution process is actually blind. Since Wiggins (1978) introduced the entirely innovative concept based on the maximization of the kurtosis, scholars have successively presented a series of blind deconvolution methods only using the characteristics of seismic records, which recover the reflectivity sequence and extract the wavelet without any additional priori assumptions.The study on blind deconvolution in seismic signal processing region is much earlier than independent component analysis (ICA) which originated as a type of new signal processing technology from blind source separation (BSS) in 1990s. Along with the maturity of ICA system info and its promising applications in image processing, biomedical signal processing, communication system, finance system, speech signal processing, radar and sonar systems etc, more and more researchers begin to merge ICA into geophysical signal analysis.This paper makes a preliminary discussion on ICA-based seismic blind deconvolution method and its application. According to the principle and algorithm of blind deconvolution methods, we classify them into five groups, also point out the two research hotspots of seismic blind deconvolution—sparse blind deconvolution based on Bayes theory and ICA-based seismic blind deconvolution.Upon analyzing model and real data, we find out the modified Cauchy distribution prior information can keep a better balance of improving the resolution of seismic data and reducing suppression of weak reflections. Combining the Bussgang-type deconvolution in ICA algorithm and seismology convolution model, we propose a negative entropy-based Bussgang seismic blind deconvolution which is better suitable for non-minimum phase and non-gaussian system and could obtain the optimum evaluation of raw reflection coefficient and be characteristic of fast convergence and high precision in the results of numeric simulation and practical data processing.Finally, we further discuss the innovative applications of ICA in the field of seismic signal processing from the practical production, which include multi-scale frequency expanding technology based on blind deconvolution, the framework of seismic ICA serial process aimed to perform better precision in low SNR data, and blind source seismic attribute optimize method taking the advantages of ICA in the aspect of data dimension reduction and feature extraction.