Convolutional neural network seismic signal denoising method based on attention guidance

Abstract

The invention provides a convolutional neural network seismic signal denoising method based on attention guidance. The method comprises the following steps: synthesizing effective signals in a seismic record by adopting Ricker wavelets; preprocessing the synthesized seismic data set, and constructing a training data set containing a noise set and a signal set; and inputting training data into an ADNet network model composed of four modules including a sparse block (SB), a feature enhancement block (FEB), an attention block (AB) and a reconstruction block (RB) for training, and after training is completed, suppressing the noise of seismic signals by using the ADNet network. According to the denoising method provided by the invention, the noise in the seismic signal can be suppressed, detail information is reserved, and the processing effect is relatively good.

Description

technical field [0001] The present invention relates to the technical field of seismic signal processing, in particular to an Attention-guided CNN for seismic signal denoising method for suppressing seismic data noise. Background technique [0002] In seismic exploration, the existence of noise will greatly affect the quality of seismic data. With the reduction of resources, the signal-to-noise ratio of the collected seismic data will decrease, and the noise properties will become more complex, especially the random noise in some areas has low-frequency, non-Gaussian , non-stationary, and high-energy characteristics, there is serious aliasing between the effective signal and random noise in the frequency domain. Therefore, in order to obtain high-quality seismic signals, it is necessary to denoise the seismic signals while maintaining the original information as much as possible. Traditional noise removal methods include: polynomial fitting method, Curvelet transform, wavel...

AI Technical Summary

Problems solved by technology

DCAENN based on self-encoding has no generalization ability; DnCNN based on residual convolutional neural network can handle Gaussian noise, but cannot remove unknown types of noise, and with the increase of network depth, the influence of shallow network on deep network will weaken ; Based on the SGAN of the generated confrontation network, the network is not easy to train, and the feature data is easy to lose during denoising

Images