Self-supervised learning with cross-training between convolutional neural network and transformer for low-frequency enhancements

A system for seismic data processing, the system comprising:

• at least one seismic sensor configured to generate seismic data associated with a surface formation;
• at least one processor configured to:
• receive the seismic data associated with the subsurface formation;
• train a convolutional neural network and a transformer neural network in a self-supervised cross-training framework, wherein to train the convolutional neural network and the transformer neural network, the at least one processor is configure to:
• generate training pairs by applying high-pass filtering to the seismic data;
• generate, using the training pairs, a first prediction from the convolutional neural network and a second prediction from the transformer neural network;
• and
• establish bidirectional flow of outputs between the convolutional neural network and the transformer neural network;
• and
• apply the trained transformer neural network to the seismic data to generate enhanced seismic data.

A non-transitory computer readable medium storing computer-executable code for enhancing seismic data, the computer-executable code comprising:

• code for receiving seismic data associated with a subsurface formation;
• code for training a convolutional neural network and a transformer neural network in a self-supervised cross-training framework, wherein the code for training includes:
• code for generating training pairs by applying high-pass filtering to the seismic data;
• code for generating, using the training pairs, a first prediction from the convolutional neural network and a second prediction from the transformer neural network;
• and
• code for establishing bidirectional flow of outputs between the convolutional neural network and the transformer neural network;
• and
• code for applying the trained transformer neural network to the seismic data to generate enhanced seismic data.