Seismic denoising based on modified BP neural network

Abstract

A new method for seismic random noise reduction based on robust function and back propagation (BP) neural network is proposed in this paper. This method introduces BP neural network utilizing least mean log squares (LMLS) error function or least trimmed squares (LTS) estimator instead of least mean squares (LMS) error function as its error function. The proposed method can diminish the influence of random noise on the accuracy of BP neural network model and improve the denoising capability of neural network, obviously. Experimental results demonstrate that the proposed new method can reduce random noise on seismic data and preserve in-phase axes more effectively than some traditional denoising methods and generic BP neural network model.