Machine learning based signal recovery
Various aspects described herein relate to a machine learning based signal recovery. In one example, a computer-implemented method of noise contaminated signal recovery includes receiving, at a server, a first signal including a first portion and a second portion, the first portion indicative of data collected by a plurality of sensors, the second portion representing noise; performing a first denoising process on the first signal to filter out the noise to yield a first denoised signal; applying a machine learning model to determine a residual signal indicative of a difference between the first signal and the first denoised signal; and determining a second signal by adding the residual signal to the first denoised signal, the second signal comprising (i) signals of the first portion with higher magnitudes than the noise in the second portion, and (ii) signals of the first portion having lower magnitudes than the noise in the second portion.
1. A computer-implemented method of noise contaminated signal recovery, the method comprising:
receiving, at a server, a first signal including a first portion and a second portion, the first portion indicative of data collected by a plurality of sensors, the second portion representing noise;
performing a first denoising process on the first signal to remove the noise to yield a first denoised signal;
applying a machine learning model to determine a residual signal indicative of a difference between the first signal and the first denoised signal; and
determining a second signal by adding the residual signal to the first denoised signal, the second signal comprising (i) signals of the first portion with higher magnitudes than the noise in the second portion, and (ii) signals of the first portion having lower magnitudes than the noise in the second portion.
2. The computer-implemented method of claim 1 , wherein the data corresponds to seismic data collected by the plurality of sensors.
3. The computer-implemented method of claim 1 , further comprising:
training the machine learning model with the first denoised signal every time the server receives a different instance of the first signal and determines a corresponding first denoised signal.
4. The computer-implemented method of claim 3 , wherein training the machine learning model includes determining a solution to an optimization problem using a k-singular value decomposition algorithm.
5. The computer-implemented method of claim 3 , wherein the training the machine learning model is unsupervised.
6. The computer-implemented method of claim 1 , wherein determining the residual signal includes determining a solution to a dual domain sparse inversion problem using a sparse representation of the residual signal and the noise.
7. The computer-implemented method of claim 6 , wherein solving the dual domain sparse inversion problem includes using a deterministic algorithm, the deterministic algorithm corresponding to one of a nonmonotone alternating direction method, or stochastic algorithm such as matching pursuit.
8. The method of claim 1 , wherein determining the second signal is based on a single dictionary representing the noise trained using the machine learning model.
9. The method of claim 1 , wherein determining the second signal is based on a dual dictionary representing the noise and the data, the dual dictionary being trained using the machine learning model.
10. A system for noise contaminated signal recovery, the system comprising:
memory having computer-readable instruction stored therein; and
one or more processors configured to execute the computer-readable instructions to:
receive a signal contaminated by noise, the signal including data collected by a plurality of sensors; and
process the signal using a machine learning model to yield a processed signal such that (1) the noise is removed from the processed signal and (2) portions of the data with magnitudes lower than magnitudes of the noise are preserved in the processed signal after removal of the noise.
11. The system of claim 10 , wherein the one or more processors are configured to execute the computer-readable instructions to process the signal by:
performing a denoising process on the signal to remove the noise to yield a denoised signal; applying the machine learning model to determine a residual signal indicative of a difference between the signal and the denoised signal; and determining a processed signal by adding the residual signal to the denoised signal.
12. The system of claim 11 , wherein the one or more processors are configured to control the plurality of sensors to determine the residual signal by determining a solution to a dual domain sparse inversion problem using a sparse representation of the residual signal and the noise.
13. The system of claim 11 , wherein the one or more processors are configured to execute the computer-readable instructions to:
train the machine learning model with the denoised signal every time a different instance of the signal is received and a corresponding denoised signal is determined.
14. The system of claim 13 , wherein the one or more processors are configured to execute the computer-readable instructions to train the machine learning model by determining a solution to an optimization problem using a k-singular value decomposition algorithm.
15. The system of claim 13 , wherein training the machine learning model is unsupervised.
16. The system of claim 10 , wherein the data corresponds to seismic data collected by the plurality of sensors.
17. The system of claim 16 , wherein the one or more processors are configured to control the plurality of sensors to collect the data over a specified period of time.
18. One or more non-transitory computer-readable medium having computer-readable instructions, which when executed by one or more processors of a system for noise contaminated signal recovery, cause the one or more processors to:
receive a signal contaminated by noise, the signal including data collected by a plurality of sensors; and
process the signal using a machine learning model to yield a processed signal such that (1) the noise is removed from the processed signal and (2) portions of the data with magnitudes lower than magnitudes of the noise are preserved in the processed signal after removal of the noise.
19. The one or more non-transitory computer-readable medium of claim 18 , wherein the execution of the computer-readable instructions by the one or more processors, cause the one or more processors to process the signal by:
performing a denoising process on the signal to remove the noise to yield a denoised signal; applying the machine learning model to determine a residual signal indicative of a difference between the signal and the denoised signal; and determining a processed signal by adding the residual signal to the denoised signal.
20. The one or more non-transitory computer-readable medium of claim 19 , wherein the execution of the computer-readable instructions by the one or more processors, cause the one or more processors to control the plurality of sensors to determine the residual signal by determining a solution to a dual domain sparse inversion problem using a sparse representation of the residual signal and the noise.
21. The one or more non-transitory computer-readable medium of claim 20 , wherein solving the dual domain sparse inversion problem includes using a deterministic algorithm, the deterministic algorithm corresponding to one of a nonmonotone alternating direction method, or stochastic algorithm such as matching pursuit.
22. The one or more non-transitory computer-readable medium of claim 21 , wherein the execution of the computer-readable instructions by the one or more processors, cause the one or more processors to form a dataset using the signal and perform the denoising process on the data set.
23. The one or more non-transitory computer-readable medium of claim 19 , wherein the execution of the computer-readable instructions by the one or more processors, cause the one or more processors to:
train the machine learning model with the denoised signal every time a different instance of the signal is received and a corresponding denoised signal is determined, by determining a solution to an optimization problem using a k-singular value decomposition algorithm.
24. The one or more non-transitory computer-readable medium of claim 18 , wherein the data corresponds to seismic data collected by the plurality of sensors over a specified period of time.