Method and system for faster seismic imaging using machine learning

1. A method, comprising:

acquiring, using a seismic surveying system, first seismic data regarding a first geological region of interest, wherein the seismic surveying system comprises a plurality of seismic receivers and a plurality of seismic sources,

wherein the plurality of seismic sources comprise a seismic vibrator configured to perform a vibroseis technique,

wherein at least one of the plurality of seismic receivers is a multi-component sensor that measures pressure waves in a plurality of spatial axes, and

wherein the plurality of seismic receivers comprises an accelerometer and a geophone;

determining, by a computer processor, a plurality of pre-processed gathers in a time domain using the first seismic data and seismic data processing operation, wherein the seismic data processing operation comprises a removal one or more surface waves from the first seismic data;

forward modeling, by the computer processor, a first seismic wavefield using a portion of the plurality of pre-processed gathers, a synthetic seismic source wavelet, and a first velocity model;

backward propagating, by the computer processor, a second seismic wavefield using the portion of the plurality of pre-processed gathers and the first velocity model;

determining, by the computer processor, one or more cross-correlation values between the first seismic wavefield and the second seismic wavefield;

generating, by the computer processor, a plurality of migrated gathers using the portion of the plurality of pre-processed gathers, a predetermined imaging condition, and the one or more cross-correlation values;

acquiring, using the seismic surveying system, legacy seismic data regarding a second geological region of interest that is different from the first geological region of interest;

training, by the computer processor, an initial model to produce a pre-trained machine-learning model using a first training operation and a training dataset comprising the legacy seismic data, wherein the pre-trained machine-learning model is trained to predict migrated seismic data, wherein the first training operation is performed based on a first machine-learning algorithm and a plurality of machine-learning epochs;

selecting, by the computer processor, a plurality of training gathers based on plurality of migrated gathers;

training, by the computer processor, the pre-trained machine-learning model to produce a trained model using a second training operation,

wherein the second training operation is performed using the plurality of training gathers, the pre-trained machine-learning model, and a second machine-learning algorithm;

generating, by the computer processor, predicted migrated seismic data in a depth domain using the trained model and a plurality of non-selected pre-processed gathers among the plurality of pre-processed gathers;

generating, by the computer processor, a seismic image of the first geological region of interest using the predicted migrated seismic data; and

determining, by the computer processor, a presence of one or more hydrocarbon deposits in the first geological region of interest using the seismic image.

2. The method of claim 1 ,

wherein at least one training gather among the plurality of training gathers comprises an input gather and an output gather,

wherein the input gather corresponds to a seismic gather in a time domain, and

wherein the output gather corresponds to a migrated gather in a depth domain.

3. The method of claim 1 ,

wherein at least one seismic gather of the plurality of pre-processed gathers is a shot gather that has one dimension based a source offset during a seismic survey and a second dimension based on a time value during the seismic survey, and

wherein the at least one seismic gather is filtered to remove multiple reflections.

4. The method of claim 1 ,

wherein the pre-trained machine-learning model is a convolutional neural network comprising at least one hidden layer, at least one activation function, a plurality of weights, and a plurality of biases, and

wherein the pre-trained machine-learning model corresponds a second velocity model that is different from the first velocity model.

5. The method of claim 1 , further comprising:

determining, using the pre-trained machine-learning model, predicted migrated seismic data based on input seismic data, wherein the input seismic data corresponds to a portion of the first seismic data regarding the first geological region of interest;

determining error data based on a mismatch between the predicted migrated seismic data and a portion of the plurality of training gathers; and

updating the pre-trained machine-learning model using the error data and the second machine-learning algorithm,

wherein the second machine-learning algorithm is a backward propagation algorithm.

6. The method of claim 1 , further comprising:

determining, using the computer processor, a presence of hydrocarbons in the first geological region of interest using the seismic image.

7. The method of claim 1 , further comprising:

generating the first velocity model using the first seismic data and a seismic inversion operation.

8. The method of claim 1 ,

wherein the seismic image is a stacked image comprising a plurality of actual migrated gathers and a plurality of predicted migrated gathers.

9. A system, comprising:

a seismic surveying system comprising a seismic source and a plurality of seismic receivers, wherein the seismic surveying system acquires first seismic data regarding a first geological region of interest,

wherein the seismic source is a seismic vibrator configured to perform a vibroseis technique,

wherein at least one of the plurality of seismic receivers is a multi-component sensor that measures pressure waves in a plurality of spatial axes,

wherein the plurality of seismic receivers comprises an accelerometer and a geophone, and

wherein the seismic surveying system further acquires legacy seismic data regarding a second geological region of interest that is different from the first geological region of interest; and

a seismic interpreter comprising a computer processor, wherein the seismic interpreter is coupled to the seismic surveying system, the seismic interpreter is configured to perform a method comprising:

obtaining, using the seismic surveying system, the first seismic data regarding the first geological region of interest and the legacy seismic data regarding the second geological region of interest;

determining a plurality of pre-processed gathers in a time domain using the first seismic data and seismic data processing operation, wherein the seismic data processing operation comprises a removal of one or more surface waves from the first seismic data;

forward modeling a first seismic wavefield using a portion of the plurality of pre-processed gathers, a synthetic seismic source wavelet, and a first velocity model;

backward propagating a second seismic wavefield using the portion of the plurality of pre-processed gathers and the first velocity model;

determining one or more cross-correlation values between the first seismic wavefield and the second seismic wavefield;

generating a plurality of migrated gathers using the portion of the plurality of pre-processed gathers, a predetermined imaging condition, and the one or more cross-correlation values;

training an initial model to produce a pre-trained machine-learning model using a first training operation and a training dataset, wherein the training dataset comprises the legacy seismic data, and wherein the first training operation is performed based on a first machine-learning algorithm and a plurality of machine-learning epochs, and wherein the pre-trained machine-learning model is trained to predict migrated seismic data;

selecting a plurality of training gathers based on the plurality of migrated gathers;

training the pre-trained machine-learning model to produce a trained model using a second training operation,

wherein the second training operation is performed using the plurality of training gathers, the pre-trained machine-learning model, and a second machine-learning algorithm;

generating predicted migrated seismic data in a depth domain using the trained model and a plurality of non-selected pre-processed gathers among the plurality of pre-processed gathers;

generating a seismic image of the first geological region of interest using the predicted migrated seismic data; and

determining a presence of one or more hydrocarbon deposits in the first geological region of interest using the seismic image.