• Journal of the Korean Institute of Telematics and Electronics
• /
• v.26 no.5
• /
• pp.79-84
• /
• 1989

Optical implementation of a quadratic associative memory model of neural networks is reported. Weighted $N^3$ interconnections between neurons are realized with an optical matrix-vector multiplier and interconnection holograms.

• The Journal of the Acoustical Society of Korea
• /
• v.17 no.6
• /
• pp.37-42
• /
• 1998

본 논문에서는 SCHMM에 적용된 관찰 확률 최대화에 의한 화자 적응 알고리즘을 제안한다. 제안된 알고리즘은 SCHMM의 관찰 확률 밀도들이 새로운 화자의 음성 특징을 잘 표현하지 못하는 경우 인식 성능이 저하되는 것을 막기 위하여, 적응 데이터의 각 특징 벡터들이 최대의 관찰 확률을 가질 수 있도록 관찰 확률 밀도를 결정하는 평균 벡터 μ와 분산 행렬 Σ를 기울기 탐색(gradient search) 알고리즘에 의해 반복적으로 적응시켜 주는 방법이다. SCHMM의 상태 천이 확률 A와 혼합 밀도 계수 C는 관찰 확률 밀도 적응 과정 을 거친 후, 적응 데이터로부터 구한 확률과 기존 확률의 가중 평균을 취하는 과정을 반복 하여 적응시켜 주었다. 제안된 화자 적응 알고리즘을 사용하여 단독음 인식 실험을 수행한 결과, 화자 적응을 수행하지 않았을 때와 비교하여 화자 독립 시스템에서는 평균 9.8%, 남 성 화자 종속 시스템에서는 평균 46.0%, 여성 화자 종속 시스템에서는 평균 52.7%의 인식 률 향상을 보였다.

• Computational Structural Engineering
• /
• v.7 no.1
• /
• pp.25-30
• /
• 1994

본 고에서는 강한 바람이나 지진에 의한 고층 구조물의 진동을 감소시키기 위하여, 능동제어장치가 효과적으로 사용될 수 있음을 보이었다. 그러나 능동제어장치를 실제 구조물에 널리 적용하기에는 아직 많은 문제점들이 남아있다. 첫째, 효과에 비해 고가의 진동제어장치가 필요하며 유지관리에도 많은 비용이 필요하다. 또한 진동제어장치가 원활히 작동하기 위해서는 계속된 에너지의 공급이 있어야 하는데, 지진과 같은 과도한 구조물의 진동시 이를 보장할 수는 없다. 그러므로 극한상황에서는 진동을 오히려 증폭시킬 수도 있다. 둘째, Control Force의 산정시 사용되는 목적함수의 가중행렬을 결정하는 정형화된 방법이 아직 존재하지 않으며 효율적인 산정을 위해서는 많은 시행착오가 필요하다. 그밖에도 복잡한 실제 구조물의 단순한 모형화에 따라서 여러 문제점들이 발생하기 쉬우며, 실제 제어시에는 시간지연효과나 측정되지 못한 거동에 의한 문제 등이 있다. 구조물의 진동 저감방안에 대한 지금까지의 연구결과로는, 일반적인 경우에 있어서는 Base Isolator와 같이 재료의 감쇠특성을 이용하는 방법이 추천되고 있으나, 현재 내진설계된 고층건물과 같은 대형 구조물의 진동제어 등의 몇몇 경우에 있어서 능동제어의 필요성이 인정되고 있다. 좀더 안정적이고 효율적인 제어기법의 개발에 의해 많은 실제 구조물에 적용 범위를 넓힐 수 있을 것으로 기대된다.

• Journal of the Korean Society of Manufacturing Technology Engineers
• /
• v.7 no.2
• /
• pp.62-68
• /
• 1998

This paper presents the application of computer vision for the purpose of determining the position of the unknown object in the plane. The presented control method is to estimate the six view parameters representing the relationship between the image plane coordinates and the real physical coordinates. The estimation of six parameters is indispensable for transforming the 2-dimensional camera coordinates to the 3-dimensional spatial coordinates. Then, the position of unknown point is estimated based on the estimated parameters depending on the cameras. The suitability of this control scheme is demonstrated experimentally by determining position of the unknown object in the plane.

• Journal of the Korean Society for Precision Engineering
• /
• v.15 no.11
• /
• pp.100-107
• /
• 1998

This paper presents the development of estimation model and control method based on the new robot vision. This proposed control method is accomplished using the sequential estimation scheme that permits placement of the rigid body in each of the two-dimensional image planes of monitoring cameras. Estimation model with six parameters is developed based on the model that generalizes known 4-axis scara robot kinematics to accommodate unknown relative camera position and orientation, etc. Based on the estimated parameters, depending on each camera the joint angle of robot is estimated by the iteration method. The method is experimentally tested in two ways, the estimation model test and a three-dimensional rigid body placement task. Three results show that control scheme used is precise and robust. This feature can open the door to a range of application of multi-axis robot such as assembly and welding.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2020.11a
• /
• pp.148-150
• /
• 2020

최근 VVC(Versatile Video Coding) 표준 완료 이후 JVET(Joint Video Experts Team)은 인공신경망 기반의 비디오 부호화를 위한 AhG(Ad-hoc Group) 구성하고 인공지능을 이용한 비디오 압축 기술들을 검증하고 있으며, MPEG(Moving Picture Experts Group)에서는 DNNVC(Deep Neural Network based Video Coding) 활동을 통해 딥러닝 기반의 차세대 비디오 부호화 표준 기술을 탐색하고 있다. 본 논문은 VVC 에 채택된 신경망 기반의 기술인 MIP(Matrix Weighted Intra Prediction)를 참조하여, MIP 모델의 학습에서 손실함수가 예측 성능에 미치는 영향을 분석한다. 즉, 예측의 왜곡(MSE)만을 고려한 경우와 예측오차의 부호화 비용도 함께 반영한 손실함수를 비교한다. 실험을 위해 HEVC(High Efficiency Video Coding) 화면내 예측 대비 평균적인 PSNR 향상 정도를 나타내는 성능 지표(��PSNR)를 정의한다. 실험결과 예측오차의 부호화 특성을 반영하는 손실함수를 이용한 학습이 MSE 만 고려한 학습 대비 ��PSNR 기준 평균 0.4dB 향상됨을 보였다.

• Geophysics and Geophysical Exploration
• /
• v.21 no.1
• /
• pp.15-25
• /
• 2018

Seismic inversion is a high-resolution tool to delineate the subsurface structures which may contain oil or gas. On the other hand, marine controlled-source electromagnetic (mCSEM) inversion can be a direct tool to indicate hydrocarbon. Thus, the joint inversion using both EM and seismic data together not only reduces the uncertainties but also takes advantage of both data simultaneously. In this paper, we have developed a simultaneous joint inversion approach for the direct estimation of reservoir petrophysical parameters, by linking electromagnetic and seismic data through rock physics model. A cross-gradient constraint is used to enhance the resolution of the inversion image and the maximum likelihood principle is applied to the relative weighting factor which controls the balance between two disparate data. By applying the developed algorithm to the synthetic model simulating the simplified gas field, we could confirm that the high-resolution images of petrophysical parameters can be obtained. However, from the other test using the synthetic model simulating an anticline reservoir, we noticed that the joint inversion produced different images depending on the model constraint used. Therefore, we modified the algorithm which has different model weighting matrix depending on the type of model parameters. Smoothness constraint and Marquardt-Levenberg constraint were applied to the water-saturation and porosity, respectively. When the improved algorithm is applied to the anticline model again, reliable porosity and water-saturation of reservoir were obtained. The inversion results indicate that the developed joint inversion algorithm can be contributed to the calculation of the accurate oil and gas reserves directly.

• KSCE Journal of Civil and Environmental Engineering Research
• /
• v.11 no.4
• /
• pp.133-141
• /
• 1991

The empirical method is used to integration adjustment for the coordinates revision of a national control point but the existing values are not to be changed or changed with small variation by suitable datum selection (for example, fixed points). This paper treats the modified weighted station parameter adjustment by quasi-observations, and the method used only variance elements of existing coordinates which is substituted for all covariance elements. The movement detection of unstable points and the junction adjustment of new networks are successfully executed by the method, in integration of new secondary networks to old-secondary-triangulation points which are in the absence of the original observations in Korea. The investigation results reveal that the accuracy of old-secondary-triangulation points is ${\pm}16^{{\prime}{\prime}}$(${\pm}0.08m$), which results from the densification of test network and the analyses of old survey specifications. and is ${\pm}2.3^{{\prime}{\prime}}$ in fixing of old-secondary-triangulation points.

• The Korean Journal of Applied Statistics
• /
• v.30 no.1
• /
• pp.1-12
• /
• 2017

Many stochastic processes satisfy the Markov property exactly or at least approximately. An interested property in the Markov process is the first passage time. Since the sequential analysis by Wald, the approximation of the first passage time has been studied extensively. The Statistical computing technique due to the development of high-speed computers made it possible to calculate the values of the properties close to the true ones. This article introduces an exponentially weighted moving average (EWMA) control chart as an example of the Markov process, and studied how to calculate the average run length with problematic issues that should be cautioned for correct calculation. The results derived for approximation of the first passage time in this research can be applied to any of the Markov processes. Especially the approximation of the continuous time Markov process to the discrete time Markov chain is useful for the studies of the properties of the stochastic process and makes computational approaches easy.

• The Journal of the Acoustical Society of Korea
• /
• v.24 no.1
• /
• pp.1-10
• /
• 2005

This paper presents a wideband LPA (local polynomial approximation) beamforming algorithm that is appropriate for wideband moving sources. The Proposed wideband LPA algorithm adopts STMV (steered minimum variance) method that utilizes a steered covariance matrix obtained from multiple frequency components in one data snapshot, instead of multiple data snapshots in one frequency bin. The wideband LPA cost function is formed using STMV weight vector. The Proposed algorithm searches for the instantaneous DOA and angular velocity that maximize the wideband LPA cost function. resulting in a higher resolution performance than that of a DS LPA beamforming algorithm. Several simulations using artificial data and sea trial data are used to demonstrate the performance of the Proposed algorithm.