US 12,112,266 B2Granted

Method for processing elastic wave data, elastic wave processing device using same, and program therefor

AssigneeIucf-Hyu (Industry-University Cooperation Foundation Hanyang University)

InventorsJoong Moo Byun, Dae Ung Yoon, Ji Ho Park

Life of this patent

Estimating the term…

Loading patent…

Abstract

Provided is a method of processing elastic wave data, the method including selecting some elastic wave traces as a first label from among a plurality of elastic wave traces received without at least some elastic wave traces missing from whole elastic wave data, training an interpolation model on a machine learning basis by using at least two or more of remaining elastic wave traces except for the first label and the first label, restoring the at least some elastic wave traces missing from the whole elastic wave data by using the trained interpolation model, training an extrapolation model on a machine learning basis by using an elastic wave trace selected as a second label from among a plurality of elastic wave traces included in whole restored elastic wave data and at least two or more of remaining elastic wave traces except for the second label, and generating an additional elastic wave trace, which have not been included in the whole elastic wave data, by using the trained extrapolation model.

Claims14

Dependent claims

A method of processing elastic wave data, the method comprising:

receiving a plurality of elastic wave traces included in whole elastic wave data, wherein at least one elastic wave trace of the whole elastic wave data is missing in the received plurality of elastic wave traces;
selecting some elastic wave traces as at least one first label from among the plurality of elastic wave traces;
training an interpolation model on a machine learning basis by using the at least one first label and at least first two or more elastic wave traces among remaining elastic wave traces except for the some elastic wave traces that are selected as the at least one first label;
restoring the whole elastic wave data including the missing at least some elastic wave traces by using the trained interpolation model,
wherein the interpolation model is configured to:
determine a first difference in phase between the at least one first label and the at least first two or more elastic wave traces, and
generate at least some elastic wave traces by applying the first difference to the restored missing at least some elastic wave traces;
training an extrapolation model on a machine learning basis by using at least one elastic wave trace selected as at least one second label from among a plurality of elastic wave traces included in the restored whole elastic wave data and at least second two or more elastic wave traces among remaining elastic wave traces in the restored whole elastic wave data except for the at least one elastic wave trace that is selected as the at least one second label;
and
generating an additional elastic wave trace, which has not been included in the whole elastic wave data, by using the trained extrapolation model, and
wherein the extrapolation model is configured to:
determine a second difference in phase between the at least one second label and the at least second two or more elastic wave traces, and
generate the additional elastic wave trace by applying the second difference.

10 more elements

An apparatus for processing elastic wave data, the apparatus comprising at least one processor configured to:

receive a plurality of elastic wave traces included in whole elastic wave data, wherein at least one elastic wave trace of the whole elastic wave data is missing in the received plurality of elastic wave traces;
select some elastic wave traces as at least one first label from among the plurality of elastic wave traces;
train an interpolation model on a machine learning basis by using the at least one first label and at least first two or more remaining elastic wave traces among remaining elastic wave traces except for the some elastic wave traces that are selected as the at least one first label;
restore the whole elastic wave data including the missing at least some elastic wave traces by using the trained interpolation model,
wherein the interpolation model is configured to:
determine a first difference in phase between the at least one first label and the at least first two or more elastic wave traces, and
generate at least some elastic wave traces by applying the first difference to the restored missing at least some elastic wave traces;
train an extrapolation model on a machine learning basis by using at least one elastic wave trace selected as at least one second label from among a plurality of elastic wave traces included in the restored whole elastic wave data and at least second two or more elastic wave traces among remaining elastic wave traces in the restored whole elastic wave data except for the at least one elastic wave trace that is selected as the at least one second label;
and
generate an additional elastic wave trace, which have not been included in the whole elastic wave data, by using the trained extrapolation model, and
wherein the extrapolation model is configured to:
determine a second difference in phase between the at least one second label and the at least second two or more elastic wave traces;
and
generate the additional elastic wave trace by applying the second difference.

11 more elements

A non-transitory computer-readable medium storing a program for managing a device thereon, wherein the program, when executed by a controller of a device, cause the device to:

receive a plurality of elastic wave traces included in whole elastic wave data, wherein at least one elastic wave trace of the whole elastic wave data is missing in the received plurality of elastic wave traces;
select some elastic wave traces as at least one first label from among the plurality of elastic wave traces;
train an interpolation model on a machine learning basis by using the at least one first label and at least first two or more elastic wave traces among remaining elastic wave traces except for the some elastic wave traces that are selected as the at least one first label;
restore the whole elastic wave data including the missing at least some elastic wave traces by using the trained interpolation model,
wherein the interpolation model is configured to:
determine a first difference in phase between the at least one first label and the at least first two or more elastic wave traces;
generate at least some elastic wave traces by applying the first difference to the restored missing at least some elastic wave traces;
train an extrapolation model on a machine learning basis by using an at least one elastic wave trace selected as at least one second label from among a plurality of elastic wave traces included in the restored whole elastic wave data and at least second two or more elastic wave traces among remaining elastic wave traces in the restored whole elastic wave data except for the at least one elastic wave trace that is selected as the at least one second label;
and
generate an additional elastic wave trace, which have not been included in the whole elastic wave data, by using the trained extrapolation model, and
wherein the extrapolation model is configured to:
determine a second difference in phase between the at least one second label and the at least second two or more elastic wave traces;
and
generate the additional elastic wave trace by applying the second difference.

11 more elements

Dependencies are inferred from the claim text.

Facts

Application: 17/264,356Filed 7 Sep 2018
Priority: KR 10-2018-008951531 Jul 2018
Published: US 2021/0304004 A130 Sep 2021
Granted: 8 Oct 2024
Term ends: Not estimatedThe source records don’t support an estimate.

Classification6 codes

G06N 3/08
Neural networks: learning methods
G06N 3/044
Architecture, e.g. interconnection topology: recurrent networks, e.g. Hopfield networks
G06N 3/0442
Recurrent networks, e.g. Hopfield networks: characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
G06N 3/09
Learning methods: supervised learning

Family

US 2021/0304004 A1Pre-grant publication · published 30 Sep 2021
US 12,112,266 B2This patent

From the US filing history. Foreign family members aren’t listed.

Timeline

Citations

Open elsewhere

Building an agent or pulling many patents? Use the API or MCP server instead of scraping pages. Get an API key