KR102236169B1 - Subsea Sediment Characteristics Data Transformation System of Multi-beam Sound Pressure Data Using Deep Neural Network - Google Patents

Subsea Sediment Characteristics Data Transformation System of Multi-beam Sound Pressure Data Using Deep Neural Network Download PDF

Info

Publication number
KR102236169B1
KR102236169B1 KR1020190169155A KR20190169155A KR102236169B1 KR 102236169 B1 KR102236169 B1 KR 102236169B1 KR 1020190169155 A KR1020190169155 A KR 1020190169155A KR 20190169155 A KR20190169155 A KR 20190169155A KR 102236169 B1 KR102236169 B1 KR 102236169B1
Authority
KR
South Korea
Prior art keywords
data
sound pressure
seabed
pressure data
characteristic
Prior art date
Application number
KR1020190169155A
Other languages
Korean (ko)
Inventor
최영진
박영민
남수용
Original Assignee
(주)지오시스템리서치
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by (주)지오시스템리서치 filed Critical (주)지오시스템리서치
Priority to KR1020190169155A priority Critical patent/KR102236169B1/en
Application granted granted Critical
Publication of KR102236169B1 publication Critical patent/KR102236169B1/en

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
    • G01V1/00—Seismology; Seismic or acoustic prospecting or detecting
    • G01V1/40—Seismology; Seismic or acoustic prospecting or detecting specially adapted for well-logging
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
    • G01V1/00—Seismology; Seismic or acoustic prospecting or detecting
    • G01V1/38—Seismology; Seismic or acoustic prospecting or detecting specially adapted for water-covered areas
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
    • G01V20/00—Geomodelling in general
    • G01V99/005—
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/60—Information retrieval; Database structures therefor; File system structures therefor of audio data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Life Sciences & Earth Sciences (AREA)
  • Geophysics (AREA)
  • General Engineering & Computer Science (AREA)
  • Geology (AREA)
  • Environmental & Geological Engineering (AREA)
  • Remote Sensing (AREA)
  • Data Mining & Analysis (AREA)
  • Acoustics & Sound (AREA)
  • Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Oceanography (AREA)
  • Multimedia (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Databases & Information Systems (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Measurement Of Velocity Or Position Using Acoustic Or Ultrasonic Waves (AREA)

Abstract

The present invention relates to a system for converting seabed sediment characteristic data of multi-beam sound pressure data by using a deep neural network and a data assimilation technique. The system comprises: a data collection unit collecting sound pressure data; a database unit converting the collected sound pressure data into two-dimensional spatial data and storing the same in a database; a seabed characteristic data conversion unit inputting the sound pressure data stored in the database into a seabed characteristic value conversion model generated based on data assimilation and deep learning, and converting the same into seabed quality characteristic values; and an output unit outputting the converted seabed quality characteristic values. The data collection unit transmits the sound pressure data collected through a multi-beam to the database unit. The database unit converts the sound pressure data into two-dimensional map information as many as the number of m x n columns so that the sound pressure data can be input to a deep learning model. According to the present invention, the system can generate seabed characteristic value data similar to the actual data through a technique for converting seabed quality characteristic values to which data assimilation and deep learning are applied, but can apply training data generated by the data assimilation technique to deep learning, thereby reducing the cost and time required to grab various places, meeting high-quality training data that was insufficient in the conventional system, and creating a model with excellent performance when creating a seabed characteristic conversion model.

Description

심층신경망과 자료동화 기법을 이용한 멀티빔 음압자료의 해저퇴적물 특성자료 변환 시스템{Subsea Sediment Characteristics Data Transformation System of Multi-beam Sound Pressure Data Using Deep Neural Network}Subsea Sediment Characteristics Data Transformation System of Multi-beam Sound Pressure Data Using Deep Neural Network}

본 발명은 심층신경망과 자료동화 기법을 이용한 멀티빔 음압자료의 해저퇴적물 특성자료 변환 시스템에 관한 것으로 더욱 상세하게는, 최적내삽법 자료동화기법을 활용하여 학습자료를 생성하고 생성된 학습자료에 대해 심층신경망의 해저특성모델로 학습하여 멀티빔으로 수집된 음압자료를 해저질특성값으로 변환하여 해양의 퇴적상을 추정하는 기술에 관한 것이다.The present invention relates to a system for converting multi-beam sound pressure data to seabed sediment characteristic data using a deep neural network and data assimilation technique, and more particularly, to generate learning data using an optimal interpolation data assimilation technique, and with respect to the generated learning data. The present invention relates to a technique for estimating the sedimentary image of the ocean by learning with the submarine characteristic model of a deep neural network and converting the sound pressure data collected by multi-beams into submarine quality characteristic values.

후방산란은 음압이 해저면에서 반사될 때, 해저면 거칠기, 입도, 및 체적산란에 영향을 받기 때문에 후방산란 음압의 변화는 해저면의 고유한 특성을 반영한다. 또한 후방산란 음압의 공간적이 변화는 해저면의 미세지형과 표층퇴적물 조직 변화에 대한 정보를 제공 한다. 표층퇴적물의 입도분포가 뚜렷이 구분될 때 즉, 자갈, 모래, 뻘이 구별되어 분포하고 있는 해저면에서 측정한 후방산란 음압강도는 이들의 경계부에서 매우 뚜렷한 변화를 보인다.Since backscattering is affected by surface roughness, particle size, and volume scattering when sound pressure is reflected from the seabed, changes in the backscattered sound pressure reflect the inherent characteristics of the seafloor. In addition, the spatial change of the backscattered sound pressure provides information on the micro-topography of the seafloor and changes in the structure of surface sediments. When the particle size distribution of surface sediments is clearly distinguished, that is, the backscattered sound pressure intensity measured at the seafloor where gravel, sand, and sand are separated and distributed, shows a very distinct change at their boundary.

현재, 후방산란 음압자료를 통한 퇴적상 분류는 음압 데이터에 대해 주성분 분석을 실시 한 후 RGA(Region Growing Algorithm) 기법을 통하여 음압자료의 해저 특성값을 변환하여 해저 표층 퇴적물을 분류하고 있다.Currently, sedimentary bed classification using backscattered sound pressure data is classifying seabed surface sediments by converting the seabed characteristic values of sound pressure data through the RGA (Region Growing Algorithm) method after performing principal component analysis on the sound pressure data.

그러나, 이러한 종래의 퇴적상 분류를 통해 취득되는 음압자료의 경우, 시시각각 변화하는 해양환경적 요인에 의해 실제 퇴적상의 분포도와는 상관성이 낮아 정규화 과정에 문제가 있다.However, in the case of the conventional sound pressure data acquired through sedimentary classification, there is a problem in the normalization process because the correlation with the actual sedimentary distribution is low due to the ever-changing marine environmental factors.

이에 본 출원인은 해저질 특성을 파악하기 위해 취득된 음압자료의 해저질특성값 변환에 있어 정확도 높은 퇴적상 추정이 가능한 시스템을 제안하고자 한다.Accordingly, the present applicant intends to propose a system capable of estimating sedimentary images with high accuracy in the conversion of the acquired sound pressure data to the seabed quality characteristics in order to understand the seabed quality characteristics.

한국공개특허 제10-2010-0130537호(2010.12.13)Korean Patent Publication No. 10-2010-0130537 (2010.12.13)

본 발명의 목적은, 최적내삽법 자료동화기법을 활용하여 학습자료 생성과 생성된 학습자료를 학습하여 생성한 심층신경망 모델로 멀티빔 음압자료를 해저질특성값으로 변환하여 해양의 퇴적상을 추정함으로써, 여러 장소를 그랩(해저질 채취)하는 비용과 시간을 절감하고, 높은 정확도의 해저특성자료를 제공하는데 있다.An object of the present invention is to estimate the sedimentary phase of the ocean by converting multi-beam sound pressure data into seabed quality characteristic values with a deep neural network model generated by learning the generated learning data and generating learning data using an optimal interpolation data assimilation technique. It is to reduce the cost and time of grabbing several places (collecting the seabed), and to provide high-accuracy seabed characteristic data.

이러한 기술적 과제를 달성하기 위한 본 발명의 심층신경망과 자료동화 기법을 이용한 멀티빔 음압자료의 해저퇴적물 특성자료 변환 시스템은, 음압자료를 수집하는 데이터 수집부; 수집된 음압자료를 2차원 공간데이터로 변환하여 데이터베이스에 저장하는 데이터베이스부; 데이터베이스에 저장된 음압자료를 자료동화와 딥러닝 기반하에 생성된 해저특성자료 변환모델에 입력하여 해저질특성값으로 변환하는 해저특성자료 변환부; 및 변환된 해저질특성값을 출력하는 출력부를 포함하되, 수집부는 멀티빔을 통해 수집한 음압자료를 데이터베이스부로 인가하고, 상기 데이터베이스부는 음압자료를 딥러닝의 해저특성자료 변환모델에 입력이 가능하도록

Figure 112020080056319-pat00001
컬럼 개수 만큼 2차원 맵정보로 변환하여 저장하는 것을 특징으로 한다.In order to achieve this technical problem, the system for converting multi-beam sound pressure data to the seabed sediment characteristic data using the deep neural network and data assimilation technique of the present invention includes: a data collection unit for collecting sound pressure data; A database unit converting the collected sound pressure data into 2D spatial data and storing it in a database; A subsea characteristic data conversion unit that inputs the sound pressure data stored in the database into a subsea characteristic data conversion model generated based on data assimilation and deep learning, and converts the data into a subsea quality characteristic value; And an output unit for outputting the converted seabed quality characteristic value, wherein the collection unit applies the sound pressure data collected through the multi-beam to the database unit, and the database unit allows the sound pressure data to be input into the seabed characteristics data conversion model of deep learning.
Figure 112020080056319-pat00001
It is characterized in that it converts and stores the 2D map information as many as the number of columns.

바람직하게는 해저특성자료 변환부는, 자료동화기법을 적용하여 음압자료와 해저질특성 자료를 학습 자료로 생성하는 학습자료 생성모듈; 및 자료동화기법을 통해 생성된 학습 자료를 이용한 해저특성자료 변환모델을 생성하는 해저특성모델 생성모듈을 포함하는 것을 특징으로 한다.
학습자료 생성모듈은, 멀티빔을 통해 수집된 음압자료를 2차원 격자정보로 변환한 결과인 배경장을 생산하고, 최적내삽법 자료동화를 이용해 학습 자료인 분석장(

Figure 112020080056319-pat00112
)을 추정하되, 상기 배경장과 관측을 통해 취득한 해저질특성 관측값의 오차 공분산을 최소자승법을 통해 산정하여 상기 분석장(
Figure 112020080056319-pat00113
)을 추정하고, 분석장(
Figure 112020080056319-pat00114
)의 추정은 수학식 1을 만족하는 것을 특징으로 한다. Preferably, the subsea characteristic data conversion unit includes: a learning data generation module for generating sound pressure data and seabed quality characteristic data as learning data by applying a data assimilation technique; And a subsea characteristic model generation module for generating a subsea characteristic data transformation model using the learning data generated through the data assimilation technique.
The learning data generation module produces a background field, which is a result of converting the sound pressure data collected through multi-beams into two-dimensional grid information, and uses the optimal interpolation method to assimilate the learning data, the analysis field (
Figure 112020080056319-pat00112
), but by calculating the error covariance of the observed values of the seabed quality characteristics acquired through the background field and observations through the least squares method, the analysis field (
Figure 112020080056319-pat00113
) And the analysis field (
Figure 112020080056319-pat00114
) Is characterized in that it satisfies Equation 1.

[수학식 1][Equation 1]

Figure 112019130619228-pat00002
Figure 112019130619228-pat00002

여기서, 분석장(

Figure 112020080056319-pat00003
)은 멀티빔에 의해 수집된 음압자료를 2차원 격자정보로 변환한 결과인 배경장(
Figure 112020080056319-pat00004
)과 관측을 통해 취득한 해저질특성 관측값(
Figure 112020080056319-pat00005
)과 모델값(
Figure 112020080056319-pat00115
)의 차이에 내삽 가중치(
Figure 112020080056319-pat00007
)를 곱하여 도출하고,
Figure 112020080056319-pat00008
는 관측지점으로부터 배경장을 보간하기 위한 관측연산자이다.Here, the analysis field (
Figure 112020080056319-pat00003
) Is the background field (
Figure 112020080056319-pat00004
) And observations of seabed quality characteristics acquired through observations (
Figure 112020080056319-pat00005
) And model value (
Figure 112020080056319-pat00115
) To the difference between the interpolation weights (
Figure 112020080056319-pat00007
) To derive,
Figure 112020080056319-pat00008
Is an observation operator to interpolate the background field from the observation point.

여기서, [수학식 1]의 분석장

Figure 112020080056319-pat00116
, 배경장
Figure 112020080056319-pat00117
, 및 관측값
Figure 112020080056319-pat00118
에 대한 각 오차
Figure 112020080056319-pat00119
,
Figure 112020080056319-pat00120
,
Figure 112020080056319-pat00121
각각은 [수학식 2]로 정의되며, [수학식 2]의 오차
Figure 112020080056319-pat00122
,
Figure 112020080056319-pat00123
,
Figure 112020080056319-pat00124
각각에 대한 오차공분산
Figure 112020080056319-pat00125
,
Figure 112020080056319-pat00126
,
Figure 112020080056319-pat00127
행렬은 [수학식 3]으로 정의되는 것을 특징으로 한다.Here, the analysis field of [Equation 1]
Figure 112020080056319-pat00116
, Background sheet
Figure 112020080056319-pat00117
, And observations
Figure 112020080056319-pat00118
Each error for
Figure 112020080056319-pat00119
,
Figure 112020080056319-pat00120
,
Figure 112020080056319-pat00121
Each is defined by [Equation 2], and the error of [Equation 2]
Figure 112020080056319-pat00122
,
Figure 112020080056319-pat00123
,
Figure 112020080056319-pat00124
Error covariance for each
Figure 112020080056319-pat00125
,
Figure 112020080056319-pat00126
,
Figure 112020080056319-pat00127
The matrix is characterized by being defined by [Equation 3].

[수학식 2][Equation 2]

Figure 112020080056319-pat00128
Figure 112020080056319-pat00128

[수학식 3][Equation 3]

Figure 112020080056319-pat00129
Figure 112020080056319-pat00129

여기서,

Figure 112020080056319-pat00013
는 임의로 정해진 중앙값인 참값을 의미하며, 이 참값
Figure 112020080056319-pat00130
과 분석장
Figure 112020080056319-pat00131
의 추정치 간의 오차
Figure 112020080056319-pat00132
에 대한 오차공분산
Figure 112020080056319-pat00015
는 분석오차공분산이고, 배경장
Figure 112020080056319-pat00133
과 참값
Figure 112020080056319-pat00134
간의 오차
Figure 112020080056319-pat00135
에 대한 오차공분산
Figure 112020080056319-pat00016
는 배경오차공분산(Background error covariance)이며, 관측값
Figure 112020080056319-pat00136
과 참값
Figure 112020080056319-pat00137
간의 오차
Figure 112020080056319-pat00138
에 대한 오차공분산
Figure 112020080056319-pat00017
는 관측오차공분산(Observational error covariance)이다.here,
Figure 112020080056319-pat00013
Means the true value, which is an arbitrarily determined median value, and this true value
Figure 112020080056319-pat00130
And analyst
Figure 112020080056319-pat00131
Error between estimates of
Figure 112020080056319-pat00132
Error covariance for
Figure 112020080056319-pat00015
Is the analysis error covariance, and the background field
Figure 112020080056319-pat00133
And true value
Figure 112020080056319-pat00134
Error between
Figure 112020080056319-pat00135
Error covariance for
Figure 112020080056319-pat00016
Is the background error covariance, and the observed value
Figure 112020080056319-pat00136
And true value
Figure 112020080056319-pat00137
Error between
Figure 112020080056319-pat00138
Error covariance for
Figure 112020080056319-pat00017
Is the observational error covariance.

상기와 같은 본 발명에 따르면, 수집된 음압자료에 대해 자료동화 기법과 딥러닝을 적용한 해저질특성값 변환모델을 통해 실제와 유사한 해저질특성값을 생성함으로써, 여러 장소를 그랩하는 비용과 시간을 절감하는 효과가 있다.According to the present invention as described above, by generating a seabed quality feature value similar to the actual seabed quality feature value through a data assimilation technique and a seabed quality feature value conversion model applying deep learning to the collected sound pressure data, the cost and time of grabbing several places can be reduced. There is an effect of saving.

본 발명에 따르면, 자료동화기법으로 생성된 학습자료를 딥러닝에 적용함으로써, 해저특성 변환모델을 생성시에 종래에 부족한 양질의 학습자료를 충족시키고 성능이 우수한 모델생성이 가능한 효과가 있다.According to the present invention, by applying the learning data generated by the data assimilation technique to deep learning, there is an effect that it is possible to create a model with excellent performance and satisfies the quality learning data, which is insufficient in the prior art, when generating a subsea characteristic transformation model.

본 발명에 따르면, 심층신경망(Deep Nueral Network)모델로 여러 환경요인을 고려할 수 있는 비선형 분류 모델을 생성해 높은 정확도의 해저특성자료를 추정할 수 있다. According to the present invention, by generating a nonlinear classification model that can take into account various environmental factors with a deep neural network model, it is possible to estimate seabed characteristic data with high accuracy.

이처럼 본 발명에 따르면, 주성분 분석을 실시한 후 RGA기법을 통해 해저질 특성을 분류함에 따라 낮은 정확도와 신뢰도가 낮은 종래의 기법과 달리 자료동화를 활용해 비용과 시간적 제약이 따르는 자료수집을 해결하고, 심층신경망 모형으로 모델을 생성함으로써, 멀티빔으로 수집되는 음압자료로 해저질특성값을 산출할 수 있는 표준화된 해저질특성값 산출시스템으로 이용할 수 있다.As described above, according to the present invention, the seabed quality characteristics are classified through the RGA technique after principal component analysis, and thus, unlike conventional techniques with low accuracy and low reliability, data collection is solved by using data assimilation, resulting in cost and time constraints. By creating a model with a deep neural network model, it can be used as a standardized seabed quality feature value calculation system that can calculate seabed quality feature values from sound pressure data collected by multi-beams.

도 1은 본 발명의 일 실시예에 따른 심층신경망과 자료동화 기법을 이용한 멀티빔 음압자료의 해저퇴적물 특성자료 변환 시스템을 도시한 블록도.
도 2는 본 발명의 일 실시예에 따른 심층신경망과 자료동화 기법을 이용한 멀티빔 음압자료의 해저퇴적물 특성자료 변환 시스템의 해저특성자료 변환부를 도시한 블록도.
도 3은 본 발명의 일 실시예에 따른 심층신경망과 자료동화 기법을 이용한 멀티빔 음압자료의 해저퇴적물 특성자료 변환 시스템의 학습자료 생성모듈을 도시한 블록도.
도 4는 본 발명의 일 실시예에 따른 심층신경망과 자료동화 기법을 이용한 멀티빔 음압자료의 해저퇴적물 특성자료 변환 시스템의 해저특성모델 생성모듈을 도시한 블록도.
도 5는 본 발명의 일 실시예에 따른 심층신경망과 자료동화 기법을 이용한 멀티빔 음압자료의 해저퇴적물 특성자료 변환 시스템의 저질특성별로 음압값을 환산한 것을 도시한 그래프.
도 6은 본 발명의 일 실시예에 따른 심층신경망과 자료동화 기법을 이용한 멀티빔 음압자료의 해저퇴적물 특성자료 변환 시스템의 해저특성모델 생성모듈의 신경망을 도시한 예시도.
도 7은 본 발명의 일 실시예에 따른 심층신경망과 자료동화 기법을 이용한 멀티빔 음압자료의 해저퇴적물 특성자료 변환 시스템의 행렬화된 이차원 자료를 도시한 예시도.
도 8은 본 발명의 일 실시예에 따른 심층신경망과 자료동화 기법을 이용한 멀티빔 음압자료의 해저퇴적물 특성자료 변환 시스템의 1차원 형태로 변환된 학습자료를 도시한 예시도.
도 9는 본 발명의 일 실시예에 따른 심층신경망과 자료동화 기법을 이용한 멀티빔 음압자료의 해저퇴적물 특성자료 변환 시스템의 실험에 의한 해저퇴적물 결과자료를 도시한 도면.
도 10은 본 발명의 일 실시예에 따른 심층신경망과 자료동화 기법을 이용한 멀티빔 음압자료의 해저퇴적물 특성자료 변환 시스템의 멀티빔을 이용해 수집된 음압자료를 도시한 도면.
도 11은 본 발명의 일 실시예에 따른 심층신경망과 자료동화 기법을 이용한 멀티빔 음압자료의 해저퇴적물 특성자료 변환 방법을 도시한 순서도.
도 12는 본 발명의 일 실시예에 따른 심층신경망과 자료동화 기법을 이용한 멀티빔 음압자료의 해저퇴적물 특성자료 변환 방법의 제S1106 단계의 세부과정을 도시한 순서도.
도 13은 본 발명의 일 실시예에 따른 심층신경망과 자료동화 기법을 이용한 멀티빔 음압자료의 해저퇴적물 특성자료 변환 방법의 제S1202 단계의 세부과정을 도시한 순서도.
도 14는 본 발명의 일 실시예에 따른 심층신경망과 자료동화 기법을 이용한 멀티빔 음압자료의 해저퇴적물 특성자료 변환 방법의 제S1204 단계의 세부과정을 도시한 순서도.1 is a block diagram showing a system for converting seabed sediment characteristic data from multi-beam sound pressure data using a deep neural network and data assimilation technique according to an embodiment of the present invention.
FIG. 2 is a block diagram showing a subsea property data conversion unit of a system for converting multi-beam sound pressure data to subsea sediment property data using a deep neural network and data assimilation technique according to an embodiment of the present invention.
3 is a block diagram showing a learning data generation module of a system for converting characteristic data of subsea sediments from multi-beam sound pressure data using a deep neural network and data assimilation technique according to an embodiment of the present invention.
FIG. 4 is a block diagram showing a subsea characteristic model generation module of a system for converting multi-beam sound pressure data to subsea sediment characteristic data using a deep neural network and data assimilation technique according to an embodiment of the present invention.
FIG. 5 is a graph showing the conversion of sound pressure values for each sediment quality characteristic of a system for converting multi-beam sound pressure data using a deep neural network and data assimilation technique according to an embodiment of the present invention.
FIG. 6 is an exemplary diagram showing a neural network of a subsea characteristic model generation module of a system for converting multi-beam sound pressure data to subsea sediment characteristic data using a deep neural network and a data assimilation technique according to an embodiment of the present invention.
7 is an exemplary diagram showing matrixed two-dimensional data of a system for converting seabed sediment characteristic data of multi-beam sound pressure data using a deep neural network and data assimilation technique according to an embodiment of the present invention.
FIG. 8 is an exemplary diagram showing learning data converted into a one-dimensional form of a system for converting multi-beam sound pressure data into a seabed sediment characteristic data using a deep neural network and a data assimilation technique according to an embodiment of the present invention.
9 is a view showing result data of seabed sediments by an experiment of a system for converting seabed sediment characteristic data from multi-beam sound pressure data using a deep neural network and data assimilation technique according to an embodiment of the present invention.
FIG. 10 is a diagram showing sound pressure data collected using a multi-beam of a system for converting multi-beam sound pressure data using a deep neural network and data assimilation technique according to an embodiment of the present invention.
11 is a flow chart showing a method for converting seabed sediment characteristic data from multi-beam sound pressure data using a deep neural network and data assimilation technique according to an embodiment of the present invention.
12 is a flow chart showing a detailed process of step S1106 of a method for converting seabed sediment characteristic data from multi-beam sound pressure data using a deep neural network and data assimilation technique according to an embodiment of the present invention.
13 is a flowchart showing a detailed process of step S1202 of a method for converting seabed sediment characteristic data from multi-beam sound pressure data using a deep neural network and data assimilation technique according to an embodiment of the present invention.
14 is a flowchart showing a detailed process of step S1204 of a method for converting seabed sediment characteristic data from multi-beam sound pressure data using a deep neural network and data assimilation technique according to an embodiment of the present invention.

본 발명의 구체적인 특징 및 이점들은 첨부도면에 의거한 다음의 상세한 설명으로 더욱 명백해질 것이다. 이에 앞서, 본 명세서 및 청구범위에 사용된 용어나 단어는 발명자가 그 자신의 발명을 가장 최선의 방법으로 설명하기 위해 용어의 개념을 적절하게 정의할 수 있다는 원칙에 입각하여 본 발명의 기술적 사상에 부합하는 의미와 개념으로 해석되어야 할 것이다. 또한, 본 발명에 관련된 공지 기능 및 그 구성에 대한 구체적인 설명이 본 발명의 요지를 불필요하게 흐릴 수 있다고 판단되는 경우에는, 그 구체적인 설명을 생략하였음에 유의해야 할 것이다.Specific features and advantages of the present invention will become more apparent from the following detailed description based on the accompanying drawings. Prior to this, terms or words used in the present specification and claims are based on the principle that the inventor can appropriately define the concept of the term in order to describe his or her invention in the best way. It should be interpreted as a corresponding meaning and concept. In addition, when it is determined that a detailed description of known functions and configurations thereof related to the present invention may unnecessarily obscure the subject matter of the present invention, it should be noted that the detailed description has been omitted.

도 1을 참조하면, 본 발명의 일 실시예에 따른 심층신경망과 자료동화 기법을 이용한 멀티빔 음압자료의 해저퇴적물 특성자료 변환 시스템(S)은, 음압자료를 수집하는 데이터 수집부(100)와, 수집된 음압자료를 2차원 공간데이터로 변환하여 데이터베이스에 저장하는 데이터베이스부(200)와, 데이터베이스에 저장된 음압자료를 자료동화와 딥러닝 기반하에 생성된 해저질특성값 변환 모델에 입력하여 해저질특성값으로 변환하는 해저특성자료 변환부(300), 및 변환된 해저질특성값을 출력하는 출력부(400)를 포함하여 구성된다.Referring to FIG. 1, a system (S) for converting seabed sediment characteristic data of multi-beam sound pressure data using a deep neural network and data assimilation technique according to an embodiment of the present invention includes a data collection unit 100 for collecting sound pressure data and , The database unit 200 converts the collected sound pressure data into two-dimensional spatial data and stores it in a database, and the sound pressure data stored in the database is input to the seabed quality characteristic value conversion model generated based on data assimilation and deep learning. It is configured to include a seabed characteristic data conversion unit 300 for converting to a characteristic value, and an output unit 400 for outputting the converted seabed quality characteristic value.

이때, 데이터 수집부(100)는 멀티빔을 통해 수집한 음압자료를 데이터베이스부(200)로 인가하고, 데이터베이스부(200)는 음압자료 정보를 딥러닝 모델에 입력이 가능하도록

Figure 112019130619228-pat00018
컬럼(column)수 만큼 2차원 맵정보로 변환하여 저장한다.At this time, the data collection unit 100 applies the sound pressure data collected through the multi-beam to the database unit 200, and the database unit 200 allows the sound pressure data information to be input into the deep learning model.
Figure 112019130619228-pat00018
Converts and stores 2D map information as many as the number of columns.

도 2를 참조하면, 해저특성자료 변환부(300)는 자료동화기법을 적용하여 음압자료와 해저질특성자료를 학습 자료로 생성하는 학습자료 생성모듈(302), 및 자료동화기법을 통해 생성된 학습 자료를 이용한 해저특성자료 변환모델을 생성하는 해저특성모델 생성모듈(304)을 포함하여 구성된다.Referring to FIG. 2, the subsea characteristic data conversion unit 300 is a learning data generation module 302 that generates sound pressure data and seabed quality characteristic data as learning data by applying a data assimilation technique, and generated through a data assimilation technique. It is configured to include a seabed feature model generation module 304 for generating a seabed feature data transformation model using the learning data.

도 3은 본 발명의 일 실시예에 따른 심층신경망과 자료동화 기법을 이용한 멀티빔 음압자료의 해저퇴적물 특성자료 변환 시스템(S)의 학습자료 생성모듈을 도시한 구성도이다.FIG. 3 is a block diagram showing a learning data generation module of a system S for converting multi-beam sound pressure data into a seabed sediment characteristic data using a deep neural network and a data assimilation technique according to an embodiment of the present invention.

학습자료 생성모듈(302)은 음압자료와 관측을 통해 취득한 관측자료를 2차원 격자정보로 변환하는데, 이는 자료동화 알고리즘에 적용하기 위해 음압값과 관측에 의해 취득된 해저질특성 관측값 각각에 위도 및 경도를 부여하기 위함이다. 이때, 학습자료 생성모듈(302)은 동일 격자사이즈의 위도 및 경도 행렬을 생성하여 위치정보를 부여하게 된다.The learning data generation module 302 converts the sound pressure data and the observation data acquired through observation into 2D grid information, which is applied to the data assimilation algorithm, and the latitude is applied to each of the sound pressure value and the observation value of the seabed quality characteristics acquired by the observation. And to impart hardness. At this time, the learning data generation module 302 generates latitude and longitude matrices of the same grid size to provide location information.

또한, 학습자료 생성모듈(302)은 위치정보를 부여한 이후, 총 격자별 (

Figure 112020080056319-pat00019
) 개의 음압자료와
Figure 112020080056319-pat00020
개의 해저퇴적물을 나타내는 관측자료인 해저질특성자료를 입력받아 자료동화 설정 및 수정 절차를 수행한다. 여기서, 해저퇴적물인 해저특성자료는 실험으로 얻은 분석 결과이기 때문에 도 5에 도시된 바와 같이 해저질특성별로 음압값으로 환산할 수 있다. 이때, 음압값은 비선형 값이므로 저질분류단계에서 중앙값을 사용하여 환산해 음압값을 생성한 뒤 이 음압값을 이용하여 자료동화시에 관측값으로 이용한다.In addition, the learning data generation module 302, after giving the location information, the total grid (
Figure 112020080056319-pat00019
) Sound pressure data and
Figure 112020080056319-pat00020
Receives seabed quality characteristic data, which is observational data representing four seabed sediments, and performs data assimilation setup and correction procedures. Here, since the seabed characteristic data, which is a seabed sediment, is an analysis result obtained by an experiment, it can be converted into a sound pressure value for each seabed quality characteristic as shown in FIG. 5. At this time, since the sound pressure value is a non-linear value, the sound pressure value is converted using the median value in the low quality classification step, and then the sound pressure value is used as an observation value in data assimilation.

이때, 학습자료 생성모듈(302)은 자료동화시 최적내삽법(Optimal Interpolation)의 계산효율 및 각 음압자료의 독립성을 주기 위해

Figure 112020080056319-pat00021
의 격자영역을 병렬연산의 프로세스 수로 나누고,
Figure 112020080056319-pat00022
와
Figure 112020080056319-pat00023
중 작은값으로 영향반경을 설정한다. 여기서
Figure 112020080056319-pat00024
와
Figure 112020080056319-pat00025
는 각각
Figure 112020080056319-pat00026
의 x,y축에 대한 음압자료의 해상도이다.At this time, the learning data generation module 302 is used to provide the calculation efficiency of optimal interpolation and independence of each sound pressure data when data assimilation.
Figure 112020080056319-pat00021
Divide the grid area of s by the number of processes of parallel operation,
Figure 112020080056319-pat00022
Wow
Figure 112020080056319-pat00023
Set the radius of influence to the smaller of the values. here
Figure 112020080056319-pat00024
Wow
Figure 112020080056319-pat00025
Are each
Figure 112020080056319-pat00026
It is the resolution of the sound pressure data on the x,y axis of.

또한, 학습자료 생성모듈(302)은 멀티빔을 통해 수집된 음압자료를 2차원 격자정보로 변환한 결과인 배경장을 생산할 수 있는데, 이때, 최적내삽법 자료동화를 이용해 수집된 음압자료로부터 변환된 2차원 배경장으로 자료동화할 수 있으며, 그 방법으로 최적내삽법을 이용한다.In addition, the learning data generation module 302 can produce a background field, which is a result of converting the sound pressure data collected through the multi-beam into 2D grid information, at this time, converting the sound pressure data collected using the optimal interpolation method data assimilation. Data can be assimilated as a two-dimensional background field, and optimal interpolation is used as a method.

여기서, 최적내삽법은 음압자료의 배경장과 관측을 통해 취득한 해저질특성관측값에 대한 오차 공분산을 최소자승법을 통해 산정하여 분석장(

Figure 112020080056319-pat00139
)의 추정치를 계산하는 방법으로 [수학식 1]과 같이 표현된다.Here, the optimal interpolation method is the analysis field (
Figure 112020080056319-pat00139
It is expressed as [Equation 1] as a method of calculating the estimated value of ).

[수학식 1][Equation 1]

Figure 112019130619228-pat00028
Figure 112019130619228-pat00028

이때, 분석장(

Figure 112020080056319-pat00029
)의 추정치는 멀티빔에 의해 수집된 음압자료를 2차원 격자정보로 변환한 결과인 배경장(
Figure 112020080056319-pat00030
)과, 해저질특성 관측값(
Figure 112020080056319-pat00031
)과 모델값(
Figure 112020080056319-pat00140
)의 차이에 내삽 가중치(
Figure 112020080056319-pat00033
)를 곱하여 도출하고,
Figure 112020080056319-pat00034
는 관측지점으로부터 배경장을 보간하기 위한 관측연산자이다.At this time, the analysis field (
Figure 112020080056319-pat00029
The estimate of) is the background field (the result of converting the sound pressure data collected by the multi-beam into 2D grid information).
Figure 112020080056319-pat00030
) And observations of seabed quality characteristics (
Figure 112020080056319-pat00031
) And model value (
Figure 112020080056319-pat00140
) To the difference of the interpolation weight (
Figure 112020080056319-pat00033
) To derive,
Figure 112020080056319-pat00034
Is an observation operator to interpolate the background field from the observation point.

여기서, [수학식 1]에서 배경장(

Figure 112020080056319-pat00141
)에 대한 모델값(
Figure 112020080056319-pat00142
)과 해저질특성 관측값(
Figure 112020080056319-pat00143
)으로부터 도출된 분석장(
Figure 112020080056319-pat00144
), 배경장(
Figure 112020080056319-pat00145
), 및 해저질특성에 대한 관측값(
Figure 112020080056319-pat00146
) 각각의 오차
Figure 112020080056319-pat00147
,
Figure 112020080056319-pat00148
,
Figure 112020080056319-pat00149
를 [수학식 2]에 의해 도출할 수 있으며, [수학식 2]는 다음과 같이 정의하였다.Here, in [Equation 1], the background field (
Figure 112020080056319-pat00141
) For the model value (
Figure 112020080056319-pat00142
) And observations of seabed quality characteristics (
Figure 112020080056319-pat00143
Analysis field derived from (
Figure 112020080056319-pat00144
), background (
Figure 112020080056319-pat00145
), and observations for seabed quality characteristics (
Figure 112020080056319-pat00146
) Each error
Figure 112020080056319-pat00147
,
Figure 112020080056319-pat00148
,
Figure 112020080056319-pat00149
Can be derived by [Equation 2], and [Equation 2] is defined as follows.

[수학식 2][Equation 2]

Figure 112020080056319-pat00150
Figure 112020080056319-pat00150

여기서,

Figure 112020080056319-pat00038
는 임의로 정해진 중앙값인 참값을 의미하며, 이 참값
Figure 112020080056319-pat00151
과 분석장(
Figure 112020080056319-pat00152
)의 추정치 간의 오차
Figure 112020080056319-pat00153
, 배경장(
Figure 112020080056319-pat00154
) 및 해저질 특성 관측값(
Figure 112020080056319-pat00155
)의 오차
Figure 112020080056319-pat00156
와 참값
Figure 112020080056319-pat00157
간의 오차
Figure 112020080056319-pat00158
로 각각 정의하며, 이 3가지 오차의 오차공분산 행렬은 하기 [수학식 3]으로 정의한다.here,
Figure 112020080056319-pat00038
Means the true value, which is an arbitrarily determined median value, and this true value
Figure 112020080056319-pat00151
Department of Analysis (
Figure 112020080056319-pat00152
) Between estimates of
Figure 112020080056319-pat00153
, Background (
Figure 112020080056319-pat00154
) And observations of seabed quality properties (
Figure 112020080056319-pat00155
) Of the error
Figure 112020080056319-pat00156
And true value
Figure 112020080056319-pat00157
Error between
Figure 112020080056319-pat00158
Each is defined as, and the error covariance matrix of these three errors is defined by the following [Equation 3].

[수학식 3][Equation 3]

Figure 112020080056319-pat00159
Figure 112020080056319-pat00159

여기서,

Figure 112020080056319-pat00041
는 분석오차공분산이고,
Figure 112020080056319-pat00042
는 배경오차공분산(Background error covariance)이다.
Figure 112020080056319-pat00043
는 관측오차공분산(Observational error covariance)이며, [수학식 1]의
Figure 112020080056319-pat00044
는 [수학식 4]로 정의한다. here,
Figure 112020080056319-pat00041
Is the analysis error covariance,
Figure 112020080056319-pat00042
Is the background error covariance.
Figure 112020080056319-pat00043
Is the observational error covariance, and
Figure 112020080056319-pat00044
Is defined as [Equation 4].

[수학식 4][Equation 4]

Figure 112019130619228-pat00045
Figure 112019130619228-pat00045

이때, 분석오차공분산(

Figure 112020080056319-pat00046
)을 소화하는 최적의 내삽 가중치
Figure 112020080056319-pat00047
는 [수학식 5]로 정의한다. At this time, the analysis error covariance (
Figure 112020080056319-pat00046
) Optimal interpolation weight to digest
Figure 112020080056319-pat00047
Is defined as [Equation 5].

[수학식 5][Equation 5]

Figure 112020080056319-pat00160
Figure 112020080056319-pat00160

즉, [수학식 1]의 최적의 내삽가중치

Figure 112020080056319-pat00049
를 산출하기 위해서 [수학식 2]의 음압자료의 배경장
Figure 112020080056319-pat00161
과 관측값
Figure 112020080056319-pat00162
의 각 오차
Figure 112020080056319-pat00163
,
Figure 112020080056319-pat00164
와 [수학식 3]의 배경장
Figure 112020080056319-pat00165
의 배경오차공분산
Figure 112020080056319-pat00166
및 관측오차공분산
Figure 112020080056319-pat00167
가 최소자승법을 통해 산정되며, HT는 관측지점으로부터 배경장을 보간하기 위한 관측 연산자 H의 연산 결과에 대한 전치행렬이다. 그리고 E(*)는 *에 대한 오차 함수로 정의된다. 그리고, 배경장에 대한 오차상관관계는 등질적이며, 등방성을 가진다고 가정하며, 이런 경우, 배경장
Figure 112020080056319-pat00168
의 배경오차공분산
Figure 112020080056319-pat00169
는 두 점 간의 거리에만 의존한다. 두 점 간의 거리는 지수 형태의 가우시안 함수로 도출되며, 두 점 간의 거리
Figure 112020080056319-pat00170
에 대한 상관식은 [수학식 6]과 같다.That is, the optimal interpolation weight of [Equation 1]
Figure 112020080056319-pat00049
In order to calculate the background field of the sound pressure data in [Equation 2]
Figure 112020080056319-pat00161
And observations
Figure 112020080056319-pat00162
Angular error of
Figure 112020080056319-pat00163
,
Figure 112020080056319-pat00164
And background field of [Equation 3]
Figure 112020080056319-pat00165
Background error covariance of
Figure 112020080056319-pat00166
And observation error covariance
Figure 112020080056319-pat00167
Is calculated using the least squares method, and H T is the transposed matrix of the result of the operation of the observation operator H to interpolate the background field from the observation point. And E(*) is defined as an error function for *. And, the error correlation with respect to the background field is assumed to be homogeneous and isotropic. In this case, the background field
Figure 112020080056319-pat00168
Background error covariance of
Figure 112020080056319-pat00169
Depends only on the distance between the two points. The distance between two points is derived by an exponential Gaussian function, and the distance between two points
Figure 112020080056319-pat00170
The correlation equation for is as shown in [Equation 6].

[수학식 6][Equation 6]

Figure 112019130619228-pat00052
Figure 112019130619228-pat00052

이때, [수학식 6]에서

Figure 112020080056319-pat00171
는 비상관거리에 대한 함수로서 수평 비상관 거리와 수직 비상관 거리에 대해 주어지며,
Figure 112020080056319-pat00054
는 두점
Figure 112020080056319-pat00055
와
Figure 112020080056319-pat00056
사이 거리의 제곱이며
Figure 112020080056319-pat00057
는 배경장 오차 상관관계 규모로 정의하는데, 이는 계산격자점으로부터 자료동화 계산 모듈의 영향거리로 계산격자점에 가까울수록 내삽 가중치
Figure 112020080056319-pat00058
는 1에 근사하며, 거리가 멀수록
Figure 112020080056319-pat00059
는 0에 근사하게 구분된다.At this time, in [Equation 6]
Figure 112020080056319-pat00171
Is given for the horizontal and vertical uncorrelated distances as a function of the uncorrelated distance,
Figure 112020080056319-pat00054
Is two points
Figure 112020080056319-pat00055
Wow
Figure 112020080056319-pat00056
Is the square of the distance between
Figure 112020080056319-pat00057
Is defined as the scale of the background field error correlation, which is the influence distance of the data assimilation calculation module from the calculated grid point.
Figure 112020080056319-pat00058
Is close to 1, and the farther away
Figure 112020080056319-pat00059
Is approximately separated by zero.

본 발명의 일 실시예에서는 음압자료의 간격이 조밀하며, 2차원 딥러닝 모델 생성을 위해 해상도

Figure 112020080056319-pat00060
와
Figure 112020080056319-pat00061
중 작은값으로 영향반경을 설정한다. 또한,
Figure 112020080056319-pat00062
이
Figure 112020080056319-pat00063
와
Figure 112020080056319-pat00064
의 작은값 보다 큰 경우 해저질 특성은 음압 자료의 해상도에 대해 비종속적인 것으로 설정된다.In an embodiment of the present invention, the spacing of sound pressure data is dense, and the resolution is used to generate a two-dimensional deep learning model.
Figure 112020080056319-pat00060
Wow
Figure 112020080056319-pat00061
Set the radius of influence to the smaller of the values. Also,
Figure 112020080056319-pat00062
this
Figure 112020080056319-pat00063
Wow
Figure 112020080056319-pat00064
If it is greater than a small value of, the seabed quality characteristics are set to be independent of the resolution of the sound pressure data.

그리고, 학습자료 생성모듈(302)은 자료동화 결과산출을

Figure 112020080056319-pat00065
의 격자형태로 음압값을 저질특성 분류값으로 출력하며, 음압값은 위치정보와 매칭시켜 인덱싱되어 데이터베이스부(200)에 저장된다.And, the learning data generation module 302 performs data assimilation result calculation.
Figure 112020080056319-pat00065
The sound pressure value is output as a low quality characteristic classification value in the form of a grid of, and the sound pressure value is indexed by matching with the location information and stored in the database unit 200.

한편, 도 4는 본 발명의 일 실시예에 따른 심층신경망과 자료동화 기법을 이용한 멀티빔 음압자료의 해저퇴적물 특성자료 변환 시스템(S)의 해저특성모델 생성모듈(304)을 도시한 도면이다.On the other hand, FIG. 4 is a diagram showing a submarine characteristic model generation module 304 of a system S for converting multi-beam sound pressure data into a submarine sediment characteristic data using a deep neural network and a data assimilation technique according to an embodiment of the present invention.

해저특성모델 생성모듈(304)은 자료동화된 학습자료와 멀티빔으로부터 수집되는 음압값을 입력자료를 이용하여 해저특성자료 변환모델을 생성하는데, 데이터베이스부(200)에 저장된 자료동화된 해저질특성값으로부터 심층신경망 모형에 맞는 자료형태로 변환한다.The submarine characteristic model generation module 304 generates a submarine characteristic data conversion model using input data from the data assimilated learning data and the sound pressure value collected from the multi-beam, and the data assimilated seabed quality characteristics stored in the database unit 200 It converts from the value into a data format suitable for the deep neural network model.

도 7과 도 8을 참조하면,

Figure 112020080056319-pat00066
의 행렬화된 이차원 음압 자료를 1차원 형태의 자료로 변환하고,
Figure 112020080056319-pat00067
개의 자료수만큼 학습자료를 생산해, 심층신경망 모형의 output으로 입력한다. 여기서,
Figure 112020080056319-pat00068
와
Figure 112020080056319-pat00069
는 자료동화된 저질 특성값의 격자사이즈의 i방향과 j방향이 되며,
Figure 112020080056319-pat00070
는, 해저특성이 특정지점과 이웃하는 지점이 아닌 그 이상의 지점에서는 상관성이 없으며 독립적이기 때문에 학습능력의 저하를 방지함과 동시에 음압값과 해저질특성의 상관성이 잘 학습되도록 하는 로컬영역 격자사이즈를 의미한다. 다만,
Figure 112020080056319-pat00071
는 특정 상수로 한하지 않으나, 3 내지 7의 범위로 하여 구성된다.7 and 8,
Figure 112020080056319-pat00066
Convert the matrixed two-dimensional sound pressure data of to one-dimensional data,
Figure 112020080056319-pat00067
It produces learning data as many as the number of data and inputs it as the output of the deep neural network model. here,
Figure 112020080056319-pat00068
Wow
Figure 112020080056319-pat00069
Is the i-direction and j-direction of the grid size of the data-assisted low quality characteristic value,
Figure 112020080056319-pat00070
Is, the local area grid size that prevents deterioration of learning ability and allows the correlation between the sound pressure value and the quality of the seabed to be well learned because the seabed characteristics are not correlated and independent at points beyond the point adjacent to a specific point. it means. but,
Figure 112020080056319-pat00071
Is not limited to a specific constant, but is constituted in the range of 3 to 7.

구체적으로, 해저특성모델 생성모듈(304)의 분류모델 설계는 도 6에 도시된 바와 같이, 입력층(input layer), 은닉층(hidden layer), 및 출력층(output layer)로 구성하며, 자료의 경우 상기에서 준비된 전체 자료를 약 6:2:2의 비율로 훈련집합(training set), 검증집합(validation set), 및 시험집합(test set)로 나눈 이후 이들을 사용하여 심층신경망 모델을 확립한다.Specifically, the classification model design of the subsea characteristic model generation module 304 is composed of an input layer, a hidden layer, and an output layer, as shown in FIG. 6, and in the case of data After dividing the total data prepared above into a training set, a validation set, and a test set at a ratio of about 6:2:2, a deep neural network model is established using these.

이때, 입력층과 출력층의 경우에는 같은 개수만큼의 노드들로 구성하였고, 은닉층의 활성화 함수로는 hypertangent 함수를 사용하였으며, 그 수식은 [수학식 7]에 표기하였다. 그리고 출력층의 활성화 함수는 분류모형에 사용되는 softmax를 사용하였으며 그 수식은 [수학식 8]과 같다.At this time, in the case of the input layer and the output layer, the same number of nodes were formed, and a hypertangent function was used as the activation function of the hidden layer, and the equation was expressed in [Equation 7]. In addition, the activation function of the output layer used softmax used in the classification model, and its formula is as shown in [Equation 8].

[수학식 7][Equation 7]

Figure 112019130619228-pat00072
Figure 112019130619228-pat00072

[수학식 8][Equation 8]

Figure 112019130619228-pat00073
Figure 112019130619228-pat00073

또한, 해저특성모델 생성모듈(304)의 딥러닝 모델 학습시 학습자료에만 과하게 적응하여 일반화 성능이 떨어지는 것을 방지하기 위해 가중치 감소(weight decay)기법중 L2 Regularization을 적용한다.In addition, L2 Regularization among weight decay techniques is applied to prevent deterioration of generalization performance due to excessive adaptation to only the training data when the deep learning model of the subsea characteristic model generation module 304 is trained.

이때, [수학식 9]를 참조하면 L2 Regularization은 기존 손실함수(

Figure 112019130619228-pat00074
)에 모든 학습파라미터에 제곱을 더한 식을 새로운 손실함수로 하여 가중치변동의 크기를 제어하는 기능을 수행한다. At this time, referring to [Equation 9], L2 Regularization is the existing loss function (
Figure 112019130619228-pat00074
), plus the square of all learning parameters, as a new loss function, to control the magnitude of the weight fluctuation.

[수학식 9][Equation 9]

Figure 112019130619228-pat00075
Figure 112019130619228-pat00075

이와 같이 해저특성모델 생성모듈(304)에 의해 생성된 최적의 심층신경망 모형을 이용하여 출력부(400)가 멀티빔으로부터 수집되는 음압자료를 해저질특성값으로 결과를 출력한다. 이때, 생성되는 해저질특성값은 심층신경망 모형의 구조에 맞추어 결과를 출력하게 된다.In this way, the output unit 400 outputs a result of the sound pressure data collected from the multi-beam as a seabed quality characteristic value using the optimal deep neural network model generated by the subsea characteristic model generation module 304. At this time, the generated seabed quality characteristic values are output according to the structure of the deep neural network model.

도 8을 참조하면 심층신경망 모형을 통해 출력되는 결과들은

Figure 112019130619228-pat00076
격자지점에서 중첩되며 이는 앙상블 평균(Ensemble Average)하여 최종적으로 값을 산출한다. 앙상블 평균을 통한
Figure 112019130619228-pat00077
위치에서의 결과를 산출하는 기법은 [수학식 10]에 명시하였다.Referring to Figure 8, the results output through the deep neural network model are
Figure 112019130619228-pat00076
They are overlapped at the grid points, and the values are finally calculated by performing an ensemble average. Through ensemble mean
Figure 112019130619228-pat00077
The technique for calculating the result at the location is specified in [Equation 10].

[수학식 10][Equation 10]

Figure 112019130619228-pat00078
Figure 112019130619228-pat00078

한편, 도 9 및 도 10을 참조하면, 음압자료의 해저퇴적물 특성자료 변환 시스템의 최종 산출물과 성능을 확인할 수 있다. 도 9와 10은 각각 실험에 의한 해저퇴적물 결과자료와 멀티빔을 이용해 수집된 음압자료를 bias 필터화 한 후의 해저특성 값을 산출하여 도시한 것이며, 상기에서 생성한 심층신경망 모형을 이용하여 산출한 해저질 특성값의 결과를 도시한 것이다. 두 결과의 비교는 하기 [표 1]에 나타내었다.On the other hand, referring to FIGS. 9 and 10, it is possible to confirm the final output and performance of the system for converting sound pressure data to seabed sediment characteristic data. 9 and 10 show the result data of the seabed sediment obtained by the experiment and the sound pressure data collected using multi-beams, respectively, by calculating the values of the seabed characteristics after bias filtering, and calculated using the deep neural network model generated above. It shows the result of the seabed quality characteristic value. The comparison of the two results is shown in Table 1 below.

[표 1][Table 1]

Figure 112019130619228-pat00079
Figure 112019130619228-pat00079

이하, 도 11을 참조하여 본 발명의 일 실시예에 따른 심층신경망과 자료동화 기법을 이용한 멀티빔 음압자료의 해저퇴적물 특성자료 변환 방법에 대해 살피면 아래와 같다.Hereinafter, referring to FIG. 11, a method for converting seabed sediment characteristic data of multi-beam sound pressure data using a deep neural network and data assimilation technique according to an embodiment of the present invention will be described as follows.

먼저, 데이터 수집부(100)가 음압자료를 수집하는 음압자료 수집한다(S1102).First, the data collection unit 100 collects sound pressure data for collecting sound pressure data (S1102).

이어서, 데이터베이스부(200)가 수집된 음압자료를 2차원 공간데이터로 변환하여 데이터베이스에 저장한다(S1104).Subsequently, the database unit 200 converts the collected sound pressure data into 2D spatial data and stores it in the database (S1104).

뒤이어, 해저특성자료 변환부(300)가 데이터베이스에 저장된 음압자료를 자료동화와 딥러닝 기반하에 생성된 해저질특성값 변환 모델에 입력하여 해저질특성값으로 변환한다(S1106).Subsequently, the subsea characteristic data conversion unit 300 inputs the sound pressure data stored in the database into the subsea quality characteristic value conversion model generated based on data assimilation and deep learning, and converts it into a subsea quality characteristic value (S1106).

그리고, 출력부(400)가 변환된 해저질특성값을 출력한다(S1108).Then, the output unit 400 outputs the converted seabed quality characteristic value (S1108).

이하, 도 12를 참조하여 본 발명의 일 실시예에 따른 심층신경망과 자료동화 기법을 이용한 멀티빔 음압자료의 해저퇴적물 특성자료 변환 방법의 제S1106 단계의 세부과정에 대해 살피면 아래와 같다.Hereinafter, a detailed process of step S1106 of the method for converting seabed sediment characteristic data of multi-beam sound pressure data using a deep neural network and data assimilation technique according to an embodiment of the present invention will be described with reference to FIG. 12.

제S1104단계 이후, 해저특성자료 변환부(300)의 학습자료 생성모듈(302)이 자료동화기법을 적용하여 음압자료와 해저특성자료를 학습 자료로 생성한다(S1202).After step S1104, the learning data generation module 302 of the subsea characteristic data conversion unit 300 applies the data assimilation technique to generate sound pressure data and submarine characteristic data as learning data (S1202).

그리고, 해저특성자료 변환부(300)의 해저특성모델 생성모듈(304)이 자료동화를 통해 생성된 학습데이터를 이용한 해저특성자료 변환모델을 생성한다(S1204).Then, the seabed feature model generation module 304 of the seabed feature data conversion unit 300 generates a seabed feature data conversion model using the learning data generated through data assimilation (S1204).

이하, 도 13을 참조하여 본 발명의 일 실시예에 따른 심층신경망과 자료동화 기법을 이용한 멀티빔 음압자료의 해저퇴적물 특성자료 변환 방법의 제S1202 단계의 세부과정에 대해 살피면 아래와 같다.Hereinafter, a detailed process of step S1202 of the method for converting seabed sediment characteristic data of multi-beam sound pressure data using a deep neural network and data assimilation technique according to an embodiment of the present invention will be described with reference to FIG. 13.

제S1104단계 이후, 학습자료 생성모듈(302)이 음압자료 2차원 격자정보 변환을 수행한다(S1302).After step S1104, the learning data generation module 302 converts the two-dimensional grid information of the sound pressure data (S1302).

이어서, 학습자료 생성모듈(302)이 관측자료 2차원 결자정보 변환을 수행한다(S1304).Subsequently, the learning data generation module 302 converts the observation data to two-dimensional missing information (S1304).

뒤이어, 학습자료 생성모듈(302)이 자료동화 설정 및 자료동화를 수행한다(S1306).Subsequently, the learning material generation module 302 performs data assimilation setting and data assimilation (S1306).

그리고, 학습자료 생성모듈(302)이 자료동화 결과를 산출한다(S1308).Then, the learning data generation module 302 calculates the result of data assimilation (S1308).

이하, 도 14를 참조하여 본 발명의 일 실시예에 따른 심층신경망과 자료동화 기법을 이용한 멀티빔 음압자료의 해저퇴적물 특성자료 변환 방법의 제S1204 단계의 세부과정에 대해 살피면 아래와 같다.Hereinafter, a detailed process of step S1204 of the method for converting seabed sediment characteristic data of multi-beam sound pressure data using a deep neural network and data assimilation technique according to an embodiment of the present invention will be described with reference to FIG.

제S1202단계 이후, 해저특성모델 생성모듈(304)이 2차생산 음압자료 변환을 수행한다(S1402).After step S1202, the subsea characteristic model generation module 304 performs secondary production sound pressure data conversion (S1402).

이어서, 해저특성모델 생성모듈(304)이 분류 모델을 설계한다(S1404).Subsequently, the subsea characteristic model generation module 304 designs a classification model (S1404).

뒤이어, 해저특성모델 생성모듈(304)이 딥러닝 모델 학습을 수행한다(S1406).Subsequently, the subsea characteristic model generation module 304 performs deep learning model training (S1406).

그리고, 해저특성모델 생성모듈(304)이 변환 모델을 생성한다(S1408).Then, the subsea characteristic model generation module 304 generates a transformation model (S1408).

이처럼 본 발명에 따르면, 주성분 분석을 실시한 후 RGA기법을 통해 해저질 특성을 분류함에 따라 낮은 정확도와 신뢰도가 낮은 종래의 기법과 달리, 자료동화를 활용해 비용과 시간적 제약이 따르는 자료수집을 해결하고, 심층신경망 모형으로 모델을 생성함으로써, 멀티빔으로 수집되는 음압자료로 해저질특성값을 산출할 수 있는 표준화된 해저질특성값 산출시스템으로 이용할 수 있다.As described above, according to the present invention, by classifying seabed quality characteristics through the RGA technique after principal component analysis, unlike conventional techniques with low accuracy and low reliability, data collection with cost and time constraints is solved by using data assimilation. , By creating a model with a deep neural network model, it can be used as a standardized seabed quality property value calculation system that can calculate seabed quality property values from sound pressure data collected by multi-beams.

이상으로 본 발명의 기술적 사상을 예시하기 위한 바람직한 실시예와 관련하여 설명하고 도시하였지만, 본 발명은 이와 같이 도시되고 설명된 그대로의 구성 및 작용에만 국한되는 것이 아니며, 기술적 사상의 범주를 일탈함이 없이 본 발명에 대해 다수의 변경 및 수정이 가능함을 당업자들은 잘 이해할 수 있을 것이다. 따라서 그러한 모든 적절한 변경 및 수정과 균등 물들도 본 발명의 범위에 속하는 것으로 간주되어야 할 것이다.Although described and illustrated in connection with a preferred embodiment for illustrating the technical idea of the present invention as described above, the present invention is not limited to the configuration and operation as illustrated and described as described above, and deviates from the scope of the technical idea. It will be appreciated by those skilled in the art that many changes and modifications can be made to the present invention without. Accordingly, all such appropriate changes and modifications and equivalents should be considered to be within the scope of the present invention.

S: 심층신경망과 자료동화 기법을 이용한 멀티빔 음압자료의 해저퇴적물 특성자료 변환 시스템
100: 데이터 수집부
200: 데이터베이스부
300: 해저특성자료 변환부
302: 학습자료 생성모듈
304: 해저특성모델 생성모듈
400: 출력부S: A system for converting multi-beam sound pressure data to seabed sediment characteristics using a deep neural network and data assimilation technique
100: data collection unit
200: database unit
300: submarine characteristic data conversion unit
302: learning material generation module
304: submarine characteristic model generation module
400: output

Claims (4)

음압자료를 수집하는 데이터 수집부;
수집된 음압자료를 2차원 공간데이터로 변환하여 데이터베이스에 저장하는 데이터베이스부;
데이터베이스에 저장된 음압자료를 자료동화와 딥러닝 기반하에 생성된 해저질특성값 변환 모델에 입력하여 해저질특성값으로 변환하는 해저특성자료 변환부; 및
변환된 해저질특성값을 출력하는 출력부를 포함하되,
상기 수집부는 멀티빔을 통해 수집한 음압자료를 상기 데이터베이스부로 인가하고, 상기 데이터베이스부는 음압자료 정보를 딥러닝 모델에 입력이 가능하도록
Figure 112020136023764-pat00172
컬럼 개수 만큼 2차원 맵정보로 변환하여 저장하도록 구비되고
상기 해저특성자료 변환부는,
자료동화기법을 적용하여 음압자료와 해저특성자료를 학습 자료로 생성하는 학습자료 생성모듈; 및
자료동화를 통해 생성된 학습자료를 이용하여 해저특성자료 변환모델을 생성하고 생성된 해저특정자료 변환모델을 이용하여 입력된 배경장에 대한 학습을 통해 최적의 분석장을 출력하는 해저특성모델 생성모듈을 포함하며,
멀티빔을 통해 수집된 음압자료로부터 2차원 격자정보로 변환한 배경장을 생산하고, 최적내삽법 자료동화를 이용해 학습 자료인 분석장(
Figure 112020136023764-pat00208
)을 추정하되,
상기 생산된 배경장과 관측을 통해 취득한 해저질특성 관측값의 오차 공분산을 최소자승법을 통해 산정하여 상기 분석장(
Figure 112020136023764-pat00209
)을 추정하고, 분석장(
Figure 112020136023764-pat00210
)의 추정은 수학식 1을 만족하는 것을 특징으로 하는 심층신경망과 자료동화 기법을 이용한 멀티빔 음압자료의 해저퇴적물 특성자료 변환 시스템.
[수학식 1]
Figure 112020136023764-pat00211

여기서, 분석장(
Figure 112020136023764-pat00212
)은 멀티빔에 의해 수집된 음압자료를 2차원 격자정보로 변환한 결과인 배경장(
Figure 112020136023764-pat00213
)에 관측을 통해 취득한 해저질 특성의 관측값(
Figure 112020136023764-pat00214
)과 모델값(
Figure 112020136023764-pat00215
)의 차이에 내삽 가중치(
Figure 112020136023764-pat00216
)를 곱하여 도출하고,
Figure 112020136023764-pat00217
는 관측지점으로부터 배경장을 보간하기 위한 관측연산자이다.
A data collection unit for collecting sound pressure data;
A database unit converting the collected sound pressure data into 2D spatial data and storing it in a database;
A seabed feature data conversion unit that inputs the sound pressure data stored in the database into a seabed quality feature value conversion model generated based on data assimilation and deep learning, and converts it into a seabed quality feature value; And
Including an output unit for outputting the converted seabed quality characteristic value,
The collection unit applies the sound pressure data collected through the multi-beam to the database unit, and the database unit allows the sound pressure data information to be input into the deep learning model.
Figure 112020136023764-pat00172
It is provided to convert and store 2D map information as many as the number of columns.
The subsea characteristic data conversion unit,
A learning data generation module that generates sound pressure data and seabed characteristic data as learning data by applying a data assimilation technique; And
A submarine feature model generation module that generates a seabed feature data transformation model using the learning data generated through data assimilation, and outputs the optimal analysis field through learning about the input background field using the generated seabed-specific data conversion model. Including,
The background field converted into 2D grid information from the sound pressure data collected through the multi-beam is produced, and the analysis field, which is a learning data, is used using the optimal interpolation method.
Figure 112020136023764-pat00208
), but
The analysis field (
Figure 112020136023764-pat00209
) And the analysis field (
Figure 112020136023764-pat00210
The estimation of) satisfies Equation 1, characterized in that it satisfies Equation 1, and a system for converting multi-beam sound pressure data to seabed sediment characteristic data using a deep neural network and data assimilation technique.
[Equation 1]
Figure 112020136023764-pat00211

Here, the analysis field (
Figure 112020136023764-pat00212
) Is the background field (
Figure 112020136023764-pat00213
), the observed value of the seabed quality characteristics acquired through observation (
Figure 112020136023764-pat00214
) And model value (
Figure 112020136023764-pat00215
) To the difference between the interpolation weights (
Figure 112020136023764-pat00216
) To derive,
Figure 112020136023764-pat00217
Is an observation operator to interpolate the background field from the observation point.
 삭제delete 삭제delete 제1항에 있어서,
[수학식 1]에서의 분석장
Figure 112020136023764-pat00183
, 배경장
Figure 112020136023764-pat00184
, 및 관측장
Figure 112020136023764-pat00185
의 각 오차
Figure 112020136023764-pat00186
,
Figure 112020136023764-pat00187
,
Figure 112020136023764-pat00188
는 [수학식 2]로 정의되며,
[수학식 2]의 오차
Figure 112020136023764-pat00189
,
Figure 112020136023764-pat00190
,
Figure 112020136023764-pat00191
각각에 대한 오차공분산
Figure 112020136023764-pat00192
,
Figure 112020136023764-pat00193
,
Figure 112020136023764-pat00194
행렬은 [수학식 3]으로 정의되는 것을 특징으로 하는 심층신경망과 자료동화 기법을 이용한 멀티빔 음압자료의 해저퇴적물 특성자료 변환 시스템.
[수학식 2]
Figure 112020136023764-pat00195

[수학식 3]
Figure 112020136023764-pat00196

여기서,
Figure 112020136023764-pat00197
는 임의로 정해진 중앙값인 참값을 의미하며, 이 참값과 분석장의 추정치간의 오차가
Figure 112020136023764-pat00198
로 정의되고, 참값
Figure 112020136023764-pat00199
과 분석장
Figure 112020136023764-pat00200
의 추정치 간의 오차에 대한 오차공분산
Figure 112020136023764-pat00201
는 분석오차공분산이고, 배경장
Figure 112020136023764-pat00202
과 참값
Figure 112020136023764-pat00203
간의 오차에 대한 오차공분산
Figure 112020136023764-pat00204
는 배경오차공분산(Background error covariance)이며, 관측값
Figure 112020136023764-pat00205
과 참값
Figure 112020136023764-pat00206
간의 오차에 대한 오차공분산
Figure 112020136023764-pat00207
는 관측오차공분산(Observational error covariance)이다.The method of claim 1,
Analysis field in [Equation 1]
Figure 112020136023764-pat00183
, Background sheet
Figure 112020136023764-pat00184
, And observation field
Figure 112020136023764-pat00185
Angular error of
Figure 112020136023764-pat00186
,
Figure 112020136023764-pat00187
,
Figure 112020136023764-pat00188
Is defined as [Equation 2],
Error of [Equation 2]
Figure 112020136023764-pat00189
,
Figure 112020136023764-pat00190
,
Figure 112020136023764-pat00191
Error covariance for each
Figure 112020136023764-pat00192
,
Figure 112020136023764-pat00193
,
Figure 112020136023764-pat00194
The matrix is defined by [Equation 3], characterized in that the deep neural network and data assimilation technique to convert multi-beam sound pressure data to the seabed sediment characteristic data conversion system.
[Equation 2]
Figure 112020136023764-pat00195

[Equation 3]
Figure 112020136023764-pat00196

here,
Figure 112020136023764-pat00197
Means the true value, which is an arbitrarily determined median, and the error between this true value and the estimated value of the analysis field is
Figure 112020136023764-pat00198
Is defined as a true value
Figure 112020136023764-pat00199
And analyst
Figure 112020136023764-pat00200
Error covariance for the error between estimates of
Figure 112020136023764-pat00201
Is the analysis error covariance, and the background field
Figure 112020136023764-pat00202
And true value
Figure 112020136023764-pat00203
Error covariance for the error between
Figure 112020136023764-pat00204
Is the background error covariance, and the observed value
Figure 112020136023764-pat00205
And true value
Figure 112020136023764-pat00206
Error covariance for the error between
Figure 112020136023764-pat00207
Is the observational error covariance.
KR1020190169155A 2019-12-17 2019-12-17 Subsea Sediment Characteristics Data Transformation System of Multi-beam Sound Pressure Data Using Deep Neural Network KR102236169B1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
KR1020190169155A KR102236169B1 (en) 2019-12-17 2019-12-17 Subsea Sediment Characteristics Data Transformation System of Multi-beam Sound Pressure Data Using Deep Neural Network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
KR1020190169155A KR102236169B1 (en) 2019-12-17 2019-12-17 Subsea Sediment Characteristics Data Transformation System of Multi-beam Sound Pressure Data Using Deep Neural Network

Publications (1)

Publication Number Publication Date
KR102236169B1 true KR102236169B1 (en) 2021-04-05

Family

ID=75461998

Family Applications (1)

Application Number Title Priority Date Filing Date
KR1020190169155A KR102236169B1 (en) 2019-12-17 2019-12-17 Subsea Sediment Characteristics Data Transformation System of Multi-beam Sound Pressure Data Using Deep Neural Network

Country Status (1)

Country Link
KR (1) KR102236169B1 (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR102630288B1 (en) * 2023-10-23 2024-01-29 한국지질자원연구원 Interpolation method of 3D seismic data using AI
KR102652357B1 (en) * 2023-09-25 2024-03-28 한국지질자원연구원 Interpolation method of 3D seismic data based on the machine learning

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2008275351A (en) * 2007-04-26 2008-11-13 Furuno Electric Co Ltd Bottom quality discrimination device and bottom quality discrimination method
KR20100130537A (en) 2009-06-03 2010-12-13 주식회사 환경과학기술 Method of calculating underwater sediment distribution database
JP2012122727A (en) * 2010-12-06 2012-06-28 Furuno Electric Co Ltd Bottom sediment determination device, ultrasonic sonar, and method and program for setting parameter
JP2013228261A (en) * 2012-04-25 2013-11-07 Arc Geo Support Co Ltd Method for estimating bottom material of water bottom

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2008275351A (en) * 2007-04-26 2008-11-13 Furuno Electric Co Ltd Bottom quality discrimination device and bottom quality discrimination method
KR20100130537A (en) 2009-06-03 2010-12-13 주식회사 환경과학기술 Method of calculating underwater sediment distribution database
JP2012122727A (en) * 2010-12-06 2012-06-28 Furuno Electric Co Ltd Bottom sediment determination device, ultrasonic sonar, and method and program for setting parameter
JP2013228261A (en) * 2012-04-25 2013-11-07 Arc Geo Support Co Ltd Method for estimating bottom material of water bottom

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR102652357B1 (en) * 2023-09-25 2024-03-28 한국지질자원연구원 Interpolation method of 3D seismic data based on the machine learning
KR102630288B1 (en) * 2023-10-23 2024-01-29 한국지질자원연구원 Interpolation method of 3D seismic data using AI

Similar Documents

Publication Publication Date Title
Hell et al. Gridding heterogeneous bathymetric data sets with stacked continuous curvature splines in tension
CN102446367B (en) Method for constructing three-dimensional terrain vector model based on multi-beam sonar submarine measurement data
Riegl et al. Detection of shallow subtidal corals from IKONOS satellite and QTC View (50, 200 kHz) single-beam sonar data (Arabian Gulf; Dubai, UAE)
KR102236169B1 (en) Subsea Sediment Characteristics Data Transformation System of Multi-beam Sound Pressure Data Using Deep Neural Network
CN103344954A (en) Submarine topography construction method based on multi-source water depth data fusion
CN103292792B (en) Actual measurement SVP reconstruction method suitable for submarine detection and pseudo-landform processing
KR102220748B1 (en) Ocean Prediction System Using Future Learning of Water Temperature Time Series Prediction Data Using Deep Learning
Smith The navigation surface: a multipurpose bathymetric database
KR20190043478A (en) System and method for generating high resolution digital elevation model
CN107679476A (en) A kind of Sea Ice Types Classification in Remote Sensing Image method
KR20130096854A (en) A system for generating seafloor 3-dimensional geospatial information using intensive filtering an edge of pipe line object and digital elevation value
Marks et al. The IHO-IOC GEBCO Cook Book.
Włodarczyk-Sielicka et al. The process of modelling the elevation surface of a coastal area using the fusion of spatial data from different sensors
CN108120981B (en) Shallow water depth radar remote sensing detection method
JP4513455B2 (en) 3D submarine topographic map display apparatus and method
Toscano et al. Auto hydro break line generation using LiDAR elevation and intensity data
Liu A new framework for accuracy assessment of LIDAR-derived digital elevation models
KR101029859B1 (en) Tidal compensation system using tidal information database
Prasad et al. Modelling and Analysis of Underwater Surface using ENC Chart Visualization
Lassak et al. Satellite Derived Bathymetry for the Islands of South Eastern Crete
ABDULRAHMAN et al. A Comparison Between Interpolation Methods For More Accurate Elevation Surface Using Gnss And Gis
Foonde Relative spatial accuracy evaluation of the shuttle radar topography mapping (SRTM15 V2. 0) dataset on the Cameroon continental shelf
Rasib et al. Extraction of Digital Terrain Model (DTM) over vegetated area in tropical rainforest using LiDAR
Abramova ARTIFACTS IN THE ARCTIC DIGITAL BATHYMETRY MODELS
Carignan et al. Digital elevation models of San Diego, California: procedures, data sources and analysis

Legal Events

Date Code Title Description
E701 Decision to grant or registration of patent right
GRNT Written decision to grant