• 한국컴퓨터그래픽스학회논문지
• /
• 제28권2호
• /
• pp.1-9
• /
• 2022

본 논문에서는 균열을 감지 할 때 필요한 데이터를 생성할 수 있는 벡터 기반 증강 기법과 이를 학습할 수 있는 합성곱 인공신경망(Convolution Neural Networks, ConvNet) 기법을 제안한다. 균열을 빠르고 정확하게 감지하는 것은 건물 붕괴와 낙하 사고를 사전에 방지할 수 있는 중요한 기술이다. 이 문제를 인공지능으로 해결하기 위해서는 대량의 데이터 확보가 필수적이지만, 실제 균열 이미지를 얻기 위한 상황은 대부분 위험하기 때문에 대량의 균열 데이터를 확보하기는 어렵다. 이런 데이터베이스 구축의 문제점은 인위적인 특정 부분에 변형을 주어 데이터의 양을 늘리는 탄성왜곡(Elastic distortion)으로 완화시킬 수 있지만, 본 논문에서는 이보다 향상된 균열 패턴 결과를 ConvNet을 활용하여 모델링한다. 탄성왜곡보다 우리의 방법이 실제 균열 패턴과 유사하게 추출된 결과를 얻을 수 있었고, 일반적인 데이터 증강에서 사용되는 픽셀 단위가 아닌, 벡터 기반으로 균열 데이터 증강을 설계함으로써 균열의 변화량 측면에서 우수한 결과를 얻을 수 있다. 결과적으로 본 논문에서는 적은 개수의 균열 데이터를 입력으로 사용했음에도 불구하고 균열의 방향 및 패턴을 다양하게 생성하여 효율적으로 균열 데이터베이스를 구축할 수 있다.

• 한국컴퓨터정보학회논문지
• /
• 제27권5호
• /
• pp.55-67
• /
• 2022

센서 데이터를 활용하여 설비의 이상 진단이 가능해졌다. 하지만 설비 이상에 대한 원인 분석은 미비한 실정이다. 본 연구에서는 센서 기반 시계열 데이터 분류 모델을 위한 해석가능한 합성곱 신경망 프레임워크를 제안한다. 연구에서 사용된 센서 기반 시계열 데이터는 실제 차량에 부착된 센서를 통해 수집되었고, 반도체의 웨이퍼 데이터는 공정 과정에서 수집되었다. 추가로 실제 기계 설비에서 수집된 주기 신호 데이터를 이용 하였으며, 충분한 학습을 위해 Data augmentation 방법론인 Scaling과 Jittering을 적용하였다. 또한, 본 연구에서는 3가지 합성곱 신경망 기반 모델들을 제안하고 각각의 성능을 비교하였다. 본 연구에서는 ResNet에 Jittering을 적용한 결과 정확도 95%, F1 점수 95%로 가장 뛰어난 성능을 보였으며, 기존 연구 대비 3%의 성능 향상을 보였다. 더 나아가 결과의 해석을 위한 XAI 방법론으로 Class Activation Map과 Layer Visualization을 제안하였으며, 센서 데이터 분류에 중요 영향을 끼치는 시계열 구간을 시각적으로 확인하였다.

• Nuclear Engineering and Technology
• /
• 제55권6호
• /
• pp.2305-2314
• /
• 2023

The governing equations of atmospheric dispersion most often taking the form of a second-order partial differential equation (PDE). Currently, typical computational codes for predicting atmospheric dispersion use the Gaussian plume model that is an analytic solution. A Gaussian model is simple and enables rapid simulations, but it can be difficult to apply to situations with complex model parameters. Recently, a method of solving PDEs using artificial neural networks called physics-informed neural network (PINN) has been proposed. The PINN assumes the latent (hidden) solution of a PDE as an arbitrary neural network model and approximates the solution by optimizing the model. Unlike a Gaussian model, the PINN is intuitive in that it does not require special assumptions and uses the original equation without modifications. In this paper, we describe an approach to atmospheric dispersion modeling using the PINN and show its applicability through simple case studies. The results are compared with analytic and fundamental numerical methods to assess the accuracy and other features. The proposed PINN approximates the solution with reasonable accuracy. Considering that its procedure is divided into training and prediction steps, the PINN also offers the advantage of rapid simulations once the training is over.

• 경영정보학연구
• /
• 제22권1호
• /
• pp.59-72
• /
• 2020

인공지능이 다루기 어려운 개념에서 아주 익숙한 도구로 자리매김 하고 있다. 이와 더불어 금융권에서도 인공지능 기술을 도입하여 기존 시스템의 문제점을 개선하고자 하는 추세이며, 그 대표적인 예가 이상거래탐지 시스템(Fraud Detection System, FDS)이다. 결제 수단의 다양화 및 전자금융거래의 증가에 따라 치밀해져 가는 사이버 금융사기(Fraud)를 기존의 규칙기반 FDS로는 탐지하기 어려워지고 있다. 이를 극복하기 위해 딥러닝 기술을 적용하여 이상거래 탐지율을 향상시키고, 이상행위에 즉각 대응하며, 탐지 결과의 반영을 자동화하고자 하는 시도가 이루어지고 있다. 딥러닝 FDS 구축에서 핵심 문제는 데이터 불균형과 이상거래 패턴의 변동이다. 본 논문에서는 생성적 적대 신경망(Generative Adversarial Network, GAN)을 활용한 오버샘플링 기법을 통해 데이터 불균형 문제를 개선하고, 이상거래 분류기로써 심층 신경망(Deep Neural Network, DNN)과 합성곱 신경망(Convolutional Neural Network, CNN)을 적용하여 이러한 문제를 개선하고자 하였다. 실험 결과, GAN 오버샘플링이 이상거래 데이터의 불균형 문제를 개선하는데 효과를 보였으며, WGAN이 가장 높은 개선 효과가 있음을 확인하였다. 또한 제안 FDS 모형의 AUC가 0.9857로 랜덤포레스트 FDS 모형에 비해 약 6.5% 향상되어, 딥러닝이 이상거래 탐지에 뛰어난 성능을 가짐을 입증하였다. 더불어 딥러닝 모형 중 DNN은 CNN에 비해 오버샘플링의 효과를 더 잘 반영함을 확인하였다.

• 스마트미디어저널
• /
• 제9권3호
• /
• pp.41-45
• /
• 2020

최근 4차 산업혁명 관련 IT기술 중에서 머신러닝과 딥러닝으로 대표되는 인공지능과 사물인터넷은 다양한 연구를 통해 여러 분야에서 우리 실생활에 적용되고 있다. 본 논문에서는 사물인터넷과 객체인식 기술을 활용한 인공지능을 적용하여 의류를 분류하고자 한다. 이를 위해 이미지 데이터셋은 웹캠과 라즈베리파이를 이용하여 의류를 촬영하고, 촬영된 이미지 데이터를 전이학습된 컨벌루션 뉴럴 네트워크 인공지능망인 구글넷에 적용하였다. 의류 이미지 데이터셋은 온전한 이미지 900개와 손상이 있는 이미지 900 그리고 총 1800개를 가지고 상하의 2개의 카테고리로 분류하였다. 분류 측정 결과는 온전한 의류 이미지에서는 약 97.78%의 정확도를 보였다. 결론적으로 이러한 측정결과와 향후 더 많은 이미지 데이터의 보완을 통해 사물인터넷 기반 플랫폼상에서 인공지능망을 활용한 여타 사물들의 객체 인식에 대한 적용 가능성을 확인하였다.

• International Journal of Internet, Broadcasting and Communication
• /
• 제15권4호
• /
• pp.142-148
• /
• 2023

In Modern days, Self-driving for modern people is an absolute necessity for transportation and many other reasons. Additionally, after the outbreak of COVID-19, driving by oneself is preferred over other means of transportation for the prevention of infection. However, due to the constant exposure to stressful situations and chronic fatigue one experiences from the work or the traffic to and from it, modern drivers often drive under drowsiness which can lead to serious accidents and fatality. To address this problem, we propose a drowsy driving prevention learning model which detects a driver's state of drowsiness. Furthermore, a method to sound a warning message after drowsiness detection is also presented. This is to use MoveNet to quickly and accurately extract the keypoints of the body of the driver and Dense Neural Network(DNN) to train on real-time driving behaviors, which then immediately warns if an abnormal drowsy posture is detected. With this method, we expect reduction in traffic accident and enhancement in overall traffic safety.

• Nuclear Engineering and Technology
• /
• 제49권7호
• /
• pp.1555-1562
• /
• 2017

A loose part monitoring system is used to identify unexpected loose parts in a nuclear reactor vessel or steam generator. It is still necessary for the mass estimation of loose parts, one function of a loose part monitoring system, to develop a new method due to the high estimation error of conventional methods such as Hertz's impact theory and the frequency ratio method. The purpose of this study is to propose a mass estimation method using a Markov decision process and compare its performance with a method using an artificial neural network model proposed in a previous study. First, how to extract feature vectors using discrete cosine transform was explained. Second, Markov chains were designed with codebooks obtained from the feature vector. A 1/8-scaled mockup of the reactor vessel for OPR1000 was employed, and all used signals were obtained by impacting its surface with several solid spherical masses. Next, the performance of mass estimation by the proposed Markov model was compared with that of the artificial neural network model. Finally, it was investigated that the proposed Markov model had matching error below 20% in mass estimation. That was a similar performance to the method using an artificial neural network model and considerably improved in comparison with the conventional methods.

• Nuclear Engineering and Technology
• /
• 제51권3호
• /
• pp.807-817
• /
• 2019

Airborne Gamma Ray Spectrometry (AGRS) with its important applications such as gathering radiation information of ground surface, geochemistry measuring of the abundance of Potassium, Thorium and Uranium in outer earth layer, environmental and nuclear site surveillance has a key role in the field of nuclear science and human life. The Broyden-Fletcher-Goldfarb-Shanno (BFGS), with its advanced numerical unconstrained nonlinear optimization in collaboration with Artificial Neural Networks (ANNs) provides a noteworthy opportunity for modern AGRS. In this study a new AGRS system empowered by ANN-BFGS has been proposed and evaluated on available empirical AGRS data. To that effect different architectures of adaptive ANN-BFGS were implemented for a sort of published experimental AGRS outputs. The selected approach among of various training methods, with its low iteration cost and nondiagonal scaling allocation is a new powerful algorithm for AGRS data due to its inherent stochastic properties. Experiments were performed by different architectures and trainings, the selected scheme achieved the smallest number of epochs, the minimum Mean Square Error (MSE) and the maximum performance in compare with different types of optimization strategies and algorithms. The proposed method is capable to be implemented on a cost effective and minimum electronic equipment to present its real-time process, which will let it to be used on board a light Unmanned Aerial Vehicle (UAV). The advanced adaptation properties and models of neural network, the training of stochastic process and its implementation on DSP outstands an affordable, reliable and low cost AGRS design. The main outcome of the study shows this method increases the quality of curvature information of AGRS data while cost of the algorithm is reduced in each iteration so the proposed ANN-BFGS is a trustworthy appropriate model for Gamma-ray data reconstruction and analysis based on advanced novel artificial intelligence systems.

• Nuclear Engineering and Technology
• /
• 제54권1호
• /
• pp.269-274
• /
• 2022

An artificial neural network (ANN) that identifies radionuclides from low-count gamma spectra of a NaI scintillator is proposed. The ANN was trained and tested using simulated spectra. 14 target nuclides were considered corresponding to the requisite radionuclide library of a radionuclide identification device mentioned in IEC 62327-2017. The network shows an average identification accuracy of 98.63% on the validation dataset, with the gross counts in each spectrum N_c = 100~10000 and the signal to noise ratio SNR = 0.05-1. Most of the false predictions come from nuclides with low branching ratio and/or similar decay energies. If the N_c>1000 and SNR>0.3, which is defined as the minimum identifiable condition, the averaged identification accuracy is 99.87%. Even when the source and the detector are covered with lead bricks and the response function of the detector thus varies, the ANN which was trained using non-shielding spectra still shows high accuracy as long as the minimum identifiable condition is satisfied. Among all the considered nuclides, only the identification accuracy of ²³⁵U is seriously affected by the shielding. Identification of other nuclides shows high accuracy even the shielding condition is changed, which indicates that the ANN has good generalization performance.

• Nuclear Engineering and Technology
• /
• 제56권7호
• /
• pp.2690-2697
• /
• 2024

With advancements in machine learning technologies, artificial neural networks (ANNs) are being widely used to improve the performance of gamma-ray spectroscopy based on NaI(Tl) scintillation detectors. Typically, the performance of ANNs is evaluated using test datasets composed of actual spectra. However, the generation of such test datasets encompassing a wide range of actual spectra representing various scenarios often proves inefficient and time-consuming. Thus, instead of measuring actual spectra, we generated virtual spectra with diverse spectral features by sampling from categorical distribution functions derived from the base spectra of six radioactive isotopes: ⁵⁴Mn, ⁵⁷Co, ⁶⁰Co, ¹³⁴Cs, ¹³⁷Cs, and ²⁴¹Am. For practical applications, we determined the optimum counting time (OCT) as the point at which the change in the Kullback-Leibler divergence (ΔKLDV) values between the synthetic spectra used for training the ANN and the virtual spectra approaches zero. The accuracies of the actual spectra were significantly improved when measured up to their respective OCTs. The outcomes demonstrated that the proposed method can effectively determine the OCTs for gamma-ray spectroscopy based on ANNs without the need to measure actual spectra.