MFIEN: multi-scale feature interactive enhancement network for seismic data denoising in desert areas

Area of Science:

• Geophysics
• Seismic Data Processing
• Artificial Intelligence in Earth Sciences

Background:

• Desert seismic data acquisition faces challenges due to complex environments and geology.
• Intense background noise and signal overlap obscure seismic reflections, hindering accurate interpretation.
• Optimizing signal-to-noise ratio (SNR) is critical for effective seismic data processing.

Purpose of the Study:

• To propose a novel deep learning network for attenuating intense background noise in desert seismic data.
• To enhance the extraction of seismic reflection information and improve data quality.

Main Methods:

• Development of the Multi-scale Feature Interaction Enhancement Network (MFIEN).
• MFIEN utilizes a multi-scale feature interaction structure for effective information integration.
• Incorporation of a fusion feature enhancement module (FFEM) with dilated convolutions and varied kernel sizes to preserve seismic record structures.

Main Results:

• MFIEN demonstrated accurate suppression of intense background noise in both synthetic and field desert data.
• The network effectively recovered weak seismic signals, leading to significant data quality enhancement.
• Preservation of structural features in seismic records was achieved without altering feature map sizes.

Conclusions:

• MFIEN provides an effective solution for seismic data denoising in challenging desert environments.
• The proposed method significantly improves the signal-to-noise ratio (SNR) and aids in accurate seismic interpretation.
• This deep learning approach advances the field of seismic data processing for noisy geological conditions.