Abstract:

Seismic data usually contains random noise generated from a wide variety of unpredictable irregular factors. Independent Component Analysis (ICA) is a new multidimensional signal processing method based on high level statistics able to achieve separation of source signals in the absence of priori information. This article attempts to apply ICA to removing random noise from seismic exploration data and analyzes the assumptions and inherent uncertainties of the technique. We use an improved preprocessing algorithm to remove additive white Gaussian noise (AWGN) first, and then, by joint approximate diagonalization of eigenmatrices (JADE algorithm), further process the data to blindly separate the effective signal from the nonGaussian distributed random noise. Furthermore, the article establishes the normative similarity coefficient criteria to identify effective signal to resolve the order uncertainty problem in ICA and achieve an effective signal extraction. Simulation experiments and the actual seismic data processing experiments show that the algorithm proposed in this paper effectively removed the random noise.

Key words: Independent Component Analysis, seismic exploration data, random noise, JADE, order uncertainty