Method for nesting and connecting residual network based on multi-branch selective kernel

Abstract

The invention relates to the technical field of seismic data processing, and discloses a method for nesting and connecting a residual network based on a multi-branch selective kernel, which comprises the following processing steps of: nesting and connecting multiple layers of residual modules to obtain a residual nesting network; and adding a multi-branch selective kernel, taking the residual nested network as the input of the multi-branch selective kernel, performing feature extraction on each branch by using convolution of different scales, and calculating and outputting a final image. According to the method, nested residual connection is adopted, residual connection can be reduced, and the calculation efficiency is improved. The nested residual connection is combined with the multi-branch selective kernel, the feature map output by the nested connection residual network can be used as the input of the multi-branch selective kernel module, and the convolution kernels of different sizes are used for multi-branch fusion to obtain the feature map rich in content, so that the method adapts to complex data processing. The method has high operation efficiency and denoising performance, and can be widely applied to random noise processing of actual seismic data.

Description

technical field [0001] The invention relates to the technical field of seismic data processing, in particular to a multi-branch selective kernel nested connection residual network method. Background technique [0002] Random noise is formed by the comprehensive action of various factors, has no fixed frequency and propagation direction, and is distributed in all time and all frequency bands, so it is difficult to effectively separate it from seismic records. Traditional seismic data noise suppression methods such as wavelet transform, f-x domain filtering, curvelet transform, and Gaussian filtering are mainly based on the predictability and sparsity of seismic data, and their noise suppression effects are limited by factors such as model assumptions and parameter settings. Adaptively processing seismic data in complex areas, the denoising effect needs to be improved. [0003] In recent years, in view of the good performance of deep learning in computer vision and image proc...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to provide a method based on multi-branch selective kernel nested connection residual network, to solve the problem that traditional methods cannot adaptively deal with complex seismic data noise suppression

Images