Research On Application Of Generative Adversarial Network In Super-resolution Reconstruction Of Seismic Image

Seismic exploration is the most important method in geophysical exploration and the most effective way to solve oil and gas exploration problems.Seismic wave imaging is an important step in seismic exploration data processing.Seismic imaging can reflect the underground geological structure and lithological distribution,and is used as an important basis for locating mineral deposits(such as oil and gas,geothermal resources).With the gradual deepening of resource exploration and development,on the one hand,people’s requirements for exploration quality(accuracy and precision)are getting higher and higher,and they hope to use high-quality seismic imaging to accurately describe the geological structure and attributes.On the other hand,prospecting the geological conditions around the target is becoming more and more complex,and there are more and more factors that affect seismic wave data acquisition,which reduces the quality of seismic wave imaging data,resulting in low seismic imaging resolution,and severely restricts the development of seismic data interpretation.In order to improve the resolution of seismic imaging images,this paper proposes an image super-resolution reconstruction algorithm based on the Laplacian pyramid and Generative adversarial network.The generator of the algorithm is pyramid structure,and the progressive up sampling method is adopted to reduce the difficulty of complex data reconstruction.The discriminator is Patch GAN,which maps the input to the result matrix to promote the reconstruction of high-frequency details of the image.Experimental results on seismic data show that this method can effectively improve the resolution of seismic profiles.After reconstruction,the continuity of the events in the profile is strengthened,and the definition of geological structures such as faults is significantly increased.In addition,the algorithm has also been experimented on public data sets.Compared with other algorithms such as Lap SRN,the algorithm can obtain higher PSNR and SSIM values when reconstructing an image enlarged by 8 times,which effectively improves the reconstruction quality of large-scale factors.In the experiment of seismic image super-resolution reconstruction,it is found that the noise in the seismic data will lead to the instability of the experimental results,and single denoising is easy to cause the loss of effective information.To solve this problem,this paper proposes a seismic profile denoising and super-resolution reconstruction algorithm based on GAN.Firstly,the residual learning strategy is used to construct the denoising subnet to remove the noise interference on the basis of protecting the effective signal,and then the back projection unit is iterated to complete the high-resolution seismic profile reconstruction.The discriminator in the algorithm is a fully convolutional neural network,which uses a larger convolution kernel to extract data features,thereby enhancing the denoising and reconstruction performance of the model.Experimental results on seismic data show that the algorithm can achieve simultaneous denoising and super-resolution reconstruction,effectively improve the quality of the seismic profile,obtain a relatively ideal resolution and signal-to-noise ratio,and the reconstructed profile of geological structures such as small cracks are clearer.