Seismic resolution improving by a sequential convolutional neural network

Area of Science:

• Geophysics
• Tunnel Engineering
• Artificial Intelligence

Background:

• Thin-bed soft rock layers are a primary cause of significant tunnel deformations.
• Accurate detection of these thin beds during geological exploration is crucial for effective tunnel design and mitigation strategies.
• Conventional seismic methods lack the necessary accuracy for precise thin-bed detection.

Purpose of the Study:

• To develop a high-resolution (HR) seismic signal processing method for improved thin-bed rock detection.
• To establish a deep learning model capable of enhancing low-resolution (LR) seismic data to HR seismic data.
• To validate the effectiveness of the proposed method on practical seismic data for geological prediction.

Main Methods:

• A deep learning dataset of low-resolution (LR) and high-resolution (HR) seismic data was generated using forward modeling.
• A one-dimensional sequential convolutional neural network (1D SCNN) architecture was designed to map LR to HR seismic sequences.
• The SCNN model was trained on the prepared dataset and applied to poststack and prestack seismic data.

Main Results:

• The trained HR seismic processing model achieved high accuracy in transforming LR seismic data to HR.
• The method effectively improved seismic resolution and restored high-frequency seismic energy.
• The enhanced seismic data enabled better recognition of thin-bed rocks.

Conclusions:

• The proposed SCNN-based HR seismic processing method significantly enhances the ability to detect thin-bed rocks.
• This advanced geological prediction technique provides a reliable basis for planning and implementing measures against tunnel deformation caused by thin-bed soft rocks.
• The method demonstrates practical applicability and effectiveness on real-world seismic data.