Seismic data reconstruction method using generative adversarial network based on moment reconstruction error...

Area of Science:

• Geophysics
• Seismic Data Processing
• Machine Learning

Background:

• Seismic data acquisition often results in spatial undersampling.
• High data regularity is crucial for seismic data processing tasks like multiple removal and inversion.
• Traditional seismic data reconstruction methods have limitations due to underlying assumptions.

Purpose of the Study:

• To propose a novel generative adversarial network (GAN) based seismic data reconstruction method.
• To overcome the limitations of traditional reconstruction techniques by avoiding assumptions about data properties.
• To improve the accuracy and amplitude preservation of reconstructed seismic data.

Main Methods:

• Developed a GAN-based seismic data reconstruction method incorporating moment reconstruction error constraints.
• Modified the GAN's error function to include weighted adversarial loss and moment reconstruction loss.
• Validated the method using theoretical model data and real seismic data, analyzing interpolation errors.

Main Results:

• The proposed GAN method effectively extracts deep data features nonlinearly without prior assumptions.
• Experimental analysis demonstrated superior performance in seismic data reconstruction compared to conventional methods.
• The moment reconstruction error constraint approach showed improved amplitude preservation in reconstructed data.

Conclusions:

• The GAN-based seismic data reconstruction method with moment reconstruction error constraints offers a robust solution for undersampled seismic data.
• This approach overcomes the applicability limitations of traditional methods.
• The enhanced method provides superior reconstruction results with better amplitude fidelity.