• ICROS
• /
• v.20 no.2
• /
• pp.31-38
• /
• 2014

인간의 행동을 통해 내재된 의도를 인식하고 그 의도에 대응하는 서비스를 제공할 수 있는 능력을 로봇에게 부여하기 위한 연구의 일환으로 모방적 동기화 및 학습에 의한 인간-로봇 상호작용(Human-Robot Interaction, HRI) 시스템의 개발이 주목받고 있다. 하지만 인간이 관찰과 모방을 통해 목적을 가진 행동을 학습하는 과정은 감각 정보를 대응하는 운동 정보로 연계하고 모방 주체와 모방 대상 간의 물리적 상태의 차이를 보정하고 관측된 행동에 내재된 의도 또는 목표를 이해하는 복잡한 메커니즘 단계의 연속이기 때문에 이를 수행하기 위한 기술개발이 필요하다. 본고에서는 실제 인간이 수행하는 모방적 동기화 및 학습에 관여하는 것으로 추정되는 거울뉴런 시스템에 대하여 소개하고 이를 HRI 시스템에 활용하기 위해 개발된 선행 기술 동향을 논하고자 한다. 또한, 본 연구실에서 관련하여 진행해온 관련 연구를 통해 현재 거울 뉴런 시스템의 발전 정도와 향후 활용 방안 및 가능성을 고찰해보도록 한다.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2021.06a
• /
• pp.10-13
• /
• 2021

본 논문에서는 과다 노출된 영상을 영상 간 변환(Image-to-Image Translation)을 위해 설계된 적대적 생성 신경망(Generative Adversarial Network)을 활용하여 복원하는 연구를 수행한다. 과다 노출 복원을 위한 기존의 연구에서 과다 노출 영역 판별, 밝기 회복, 색상 보정 과정을 거치는데, 영상 내 과다 노출 영역을 판별하는 과정에서 임의로 결정하는 파라미터에 의해 복원된 영상 결과가 달라지는 한계점을 극복하기 위해 종단간(End-to-End) 신경망을 학습시켜 과다 노출 영역을 별도의 파라미터 선택과 분할된 과정 없이 한 번에 복원하는 방법을 제안한다. 영상 간 변환 신경망 학습에 필요한 과다 노출 여부로 도메인이 분할된 데이터셋은 게임 소프트웨어를 활용하여 만들어 사용하였다. 본 연구에서는 신경망이 생성한 영상이 실제로 과다 노출 영역을 탐지하여 복원하는 것을 확인하였다. 그리고 과다 노출 영역을 탐지하여 복원하는 과정을 학습 단계별로 확인함으로써 신경망이 실제로 과다 노출 복원 과정을 학습함을 보였다.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.32 no.4
• /
• pp.691-708
• /
• 2022

As the technology of machine learning and deep learning became common, it began to be applied to research on anomaly(abnormal) detection of industrial control systems. In Korea, the HAI dataset was developed and published to activate artificial intelligence research for abnormal detection of industrial control systems, and an AI contest for detecting industrial control system security threats is being conducted. Most of the anomaly detection studies have been to create a learning model with improved performance through the ensemble model method, which is applied either by modifying the existing deep learning algorithm or by applying it together with other algorithms. In this study, a study was conducted to improve the performance of anomaly detection with a post-processing method that detects abnormal data and corrects the labeling results, rather than the learning algorithm and data pre-processing process. Results It was confirmed that the results were improved by about 10% or more compared to the anomaly detection performance of the existing model.

• Transactions of the Korean Society of Mechanical Engineers
• /
• v.17 no.5
• /
• pp.1193-1201
• /
• 1993

A new deformation resistance model for hot steel strip rolling process was formulated to improve the accuracy of roll force estimation. To improve the existing deformation resistance model more precisely, a modification function was introduced in this study. For the modification function, several factors considering material and operational conditions have been investigated and the optimal modification function was determined under the principle of minimum variability. The newly formulated modification function was applied to the deformation resistance model for ultra-low carbon steel and showed improved accuracy with about 30% decrease in terms of standard deviation of predicted roll force values against measured ones.

• Journal of Convergence for Information Technology
• /
• v.11 no.8
• /
• pp.92-99
• /
• 2021

Light-scattering fine particulate matter monitors can measure particulate matter (PM) concentrations in every second and can be designed in a portable size. They can measure the concentrations of various PM sizes (PM_1.0, PM_2.5, PM_4.0 and PM₁₀) with a single sensor. They measure the number and size of particulate matters and convert them to weight per volume (concentration). These devices show a large error for asian dust. This paper proposes a scheme that compensates the PM_2.5 concenstration error for asian dust by multiple linear regression machine learning in light-scattering PM monitors. This scheme can be effective with only two or three types of PM sizes. The experimental results compare a beta-ray PM monitor of national institute of environmental research and a light-scattering PM monitor during a month. The correlation coefficient (R²) of theses two devices was 0.927 without asian dust, but it was 0.763 due to asian dust during the entire experimental period and improved to 0.944 by the proposed machine learning.

• Explosives and Blasting
• /
• v.41 no.3
• /
• pp.26-37
• /
• 2023

In order to determine the amount of explosives that can minimize the vibration generated during tunnel construction using the blasting method, it is necessary to derive the blasting vibration coefficients, K and n, by analyzing the vibration records of trial blasting in the field or under similar conditions. In this study, we aimed to develop a technique that can derive reasonable K and n when trial blasting cannot be performed. To this end, we collected full-scale trial blast data and studied how to predict the blast vibration coefficient (K, n) according to the type of explosive, center cut blasting method, rock origin and type, and rock grade using deep learning (DL). In addition, the correction value between full-scale and borehole trial blasting results was calculated to compensate for the limitations of the borehole trial blasting results and to carry out a design that aligns more closely with reality. In this study, when comparing the available explosive amount according to the borehole trial blasting result equation, the predictions from deep learning (DL) exceed 50%, and the result with the correction value is similar to other blast vibration estimation equations or about 20% more, enabling more economical design.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2023.05a
• /
• pp.99-99
• /
• 2023

기후변화는 물 관리의 가장 큰 리스크 요인이므로 물 관리 계획을 수립하는 과정에서 기후변화의 영향을 고려하는 것이 필수적이다. 기후변화에 대한 수자원 예측 관련 연구가 이루어지고 있으나, 대부분의 연구에는 수문학적 모델링이나 시뮬레이션이 동반되는데, 이 과정에는 시간과 비용이 많이 들어가며, 지역이나 연구목적에 따른 정밀한 매개변수의 보정은 전문지식이 필요하기 때문에 현업에서 연구결과를 의사결정에 활용하기에는 한계가 있다고 볼 수 있다. 이에 따라 수문학적 모델링의 입력 및 출력 결과를 딥러닝의 학습자료로 하여 수문모델을 사용하지 않아도 효율적으로 결과를 도출할 수 있는 딥러닝 기반 Surrogate 모형에 대한 연구가 이루어지고 있으나 수자원 분야에 접목된 사례는 부재한 실정이다. 따라서 이 연구를 통해 국내 유역을 대상으로 Surrogate 모형을 구축한 뒤, 그 성능을 평가하고자 한다. 이를 위한 Surrogate 모형 구축 과정은 다음과 같다. 충주댐 유역을 대상으로 과거 20년간의 강우 및 기온 자료를 수집한 뒤, 이 자료를 바탕으로 기후변화의 영향을 고려한 3,162개의 시나리오를 생성한다. 그 후 장기유출모형 IHACRES에 생성된 시나리오를 입력자료로 하여 유입량 결과를 도출하고, 이 결과를 Python코드 기반의 딥러닝 학습자료로 하여 최적 예측 결과를 도출해내는 Surrogate 모형을 생성한 뒤 기존 장기유출모형과의 성능을 비교하고자 한다. 이와 같은 Surrogate 모형은 추가적인 데이터와 매개변수의 보정 과정이 없어도 장기유출모형과 같은 결과를 짧은 시간내에 상당히 정확하게 모사할 수 있어 시간과 비용을 줄일 수 있으며, 비전문가도 쉽게 사용할 수 있다는 장점을 가진다.

• Journal of Digital Convergence
• /
• v.17 no.8
• /
• pp.205-219
• /
• 2019

As the education and training is such a virtual environment is applied to various fields, its usability is endless. However, there is an underestimation of the depth of perception in the training environment. In order to solve this problem, we tried to solve the problem by applying the top-down correction method. However, it is difficult to classify the result as a learning effect or perception change. In this study, it was confirmed that the proportion of spatial components of urine had a significant effect on the depth perception, and it was confirmed that the size perception were corrected together. In this study, we propose a correction method using spatial component and depth perception to improve the accuracy of depth perception.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2021.05a
• /
• pp.577-578
• /
• 2021

Recently, artificial intelligence systems are being used in various fields. The accuracy of the decision algorithm of artificial intelligence is greatly affected by the amount of learning and the accuracy of the learning data. In the case of the amount of learning, a large amount of data is required because it has a decisive effect on the performance of AI. In this paper, we propose a data collection system for constructing a system that analyzes future conditions and changes in infants' conditions based on the body temperature data of infants and toddlers. The proposed system is a system that collects and transmits data, and it is believed that it can minimize the resource consumption of the server system in existing big data analysis and training data construction.

• Journal of Sensor Science and Technology
• /
• v.7 no.6
• /
• pp.418-425
• /
• 1998

Stereo cameras are the most widely used sensing systems for automated machines including robots to interact with their three-dimensional(3D) working environments. The position of a target point in the 3D world coordinates can be measured by the use of stereo cameras and the camera calibration is an important preliminary step for the task. Existing camera calibration techniques can be classified into two large categories - linear and nonlinear techniques. While linear techniques are simple but somewhat inaccurate, the nonlinear ones require a modeling process to compensate for the lens distortion and a rather complicated procedure to solve the nonlinear equations. In this paper, a method employing a neural network for the calibration problem is described for tackling the problems arisen when existing techniques are applied and the results are reported. Particularly, it is shown experimentally that by utilizing the function approximation capability of multi-layer neural networks trained by the back-propagation(BP) algorithm to learn the error pattern of a linear technique, the measurement accuracy can be simply and efficiently increased.