• Proceedings of the Korea Water Resources Association Conference
• /
• 2018.05a
• /
• pp.135-135
• /
• 2018

한강의 잠수교는 평상시에는 사람과 차의 통행이 가능하나 예측수위가 5.5m일 경우, 보행자통제, 6.2m일 경우, 차량통제를 실시한다. 잠수교는 국토교통부의 홍수예보 지점은 아니지만 그 특수성으로 인해 정확한 홍수위 예측을 통해 선행시간을 확보할 필요가 있다. 일반적으로 하천 홍수위 예측을 위해서는 강우-유출 모형과 하도추적을 위한 수리모형을 결합한 모델링이 요구되나 잠수교는 하류부 조위로 인한 배수 및 상류부 팔당댐 방류량의 영향을 받아 물리적 수리 수문모형의 구축이 상당히 제약적이다. 이에 본 연구에서는 딥러닝 오픈 라이브러리인 Tensorflow 기반의 LSTM 심층신경망(Deep Neural Network) 모형을 구축하여 잠수교의 수위예측을 수행한다. LSTM 모형의 학습과 검증을 위해 2011년부터 2017년까지의 10분단위의 잠수교 수위자료, 팔당댐의 방류량과 월곶관측소의 조위자료를 수집한 후, 2011년부터 2016년까지의 자료는 신경망 학습, 2017년 자료를 이용하여 학습된 모형을 검증하였다. 민감도 분석을 통해 LSTM 모형의 최적 매개변수를 추정하고, 이를 기반으로 선행시간(lead time) 1시간, 3시간, 6시간, 9시간, 12시간, 24시간에 대한 잠수교 수위를 예측하였다. LSTM을 이용한 1~6시간 선행시간에 대한 수위예측의 경우, 모형평가 지수 NSE(Nash-Sutcliffe Efficiency)가 1시간(0.99), 3시간(0.97), 6시간(0.93)과 같이 정확도가 매우 우수한 것으로 분석되었으며, 9시간, 12시간, 24시간의 경우, 각각 0.85, 0.82, 0.74로 선행시간이 길어질수록 심층신경망의 예측능력이 저하되는 것으로 나타났다. 하천수위 또는 유량과 같은 수문시계열 분석이 목적일 경우, 종속변수에 영향을 미칠 수 있는 가용한 모든 독립변수를 데이터화하여 선행 정보를 장기적으로 기억하고, 이를 예측에 반영하는 LSTM 심층신경망 모형은 수리 수문모형 구축이 제약적인 경우, 홍수예보를 위한 활용이 가능할 것으로 판단된다.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.18 no.2
• /
• pp.104-113
• /
• 2019

The road management institute needs lots of cost to repair road surface damage. These damages are inevitable due to natural factors and aging, but maintenance technologies for efficient repair of the broken road are needed. Various technologies have been developed and applied to cope with such a demand. Recently, maintenance technology for road surface damage repair is being developed using image information collected in the form of a black box installed in a vehicle. There are various methods to extract the damaged region, however, we will discuss the image recognition technology of the deep neural network structure that is actively studied recently. In this paper, we introduce a new neural network which can estimate the road damage and its location in the image by region-based convolution neural network algorithm. In order to develop the algorithm, about 600 images were collected through actual driving. Then, learning was carried out and compared with the existing model, we developed a neural network with 10.67% accuracy.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.24 no.12
• /
• pp.1588-1594
• /
• 2020

In this paper, we discuss a data balancing method for learning a neural network that generates digital holograms using a deep neural network (DNN). Deep neural networks are based on deep learning (DL) technology and use a generative adversarial network (GAN) series. The fringe pattern, which is the basic unit of a hologram to be created through a deep neural network, has very different data types depending on the hologram plane and the position of the object. However, because the criteria for classifying the data are not clear, an imbalance in the training data may occur. The imbalance of learning data acts as a factor of instability in learning. Therefore, it presents a method for classifying and balancing data for which the classification criteria are not clear. And it shows that learning is stabilized through this.

• Annual Conference on Human and Language Technology
• /
• 2018.10a
• /
• pp.425-429
• /
• 2018

지식 그래프는 많은 수의 개채와 이들 사이의 관계를 저장하고 있기 때문에 많은 연구에서 중요한 자원으로 활용된다. 최근에는 챗봇과 질의응답과 같은 연구에서 자연어 생성을 위한 연구에 활용되고 있다. 특히 자연어 생성에서 최근 발전 된 심층 신경망이 사용되고 있는데, 이러한 방식은 모델 학습을 위한 많은 양의 데이터가 필요하다. 즉, 심층신경망을 기반으로 지식 그래프에서 문장을 생성하기 위해서는 많은 트리플과 문장 쌍 데이터가 필요하지만 학습을 위해 사용하기엔 데이터가 부족하다는 문제가 있다. 따라서 본 논문에서는 데이터 부족 문제를 해결하기 위해 핵심어 시퀀스를 추출하여 학습하는 방법을 제안하고, 학습된 모델을 통해 트리플을 입력으로 하여 자연어 문장을 생성한다. 부족한 트리플과 문장 쌍 데이터를 대체하기 위해 핵심어 시퀀스를 추출하는 모듈을 사용해 핵심어 시퀀스와 문장 쌍 데이터를 생성하였고, 순환 신경망 기반의 인코더 - 디코더 모델을 사용해 자연어 문장을 생성하였다. 실험 결과, 핵심어 시퀀스와 문장 쌍 데이터를 이용해 학습된 모델을 이용해 트리플에서 자연어 문장 생성이 원활히 가능하며, 부족한 트리플과 문장 쌍 데이터를 대체하는데 효과적임을 밝혔다.

• Journal of Broadcast Engineering
• /
• v.24 no.1
• /
• pp.87-96
• /
• 2019

Improvement of low-light image mainly focuses on the contrast enhancement. Many researches have been carried out for brightness enhancement, contrast improvement and illumination reduction. Recently, the aforementioned approaches have been replaced by artificial neural networks. This paper proposes a methodology that can replace the Retinex-based reflectance image acquisition by deep neural network. Experiments carried out on 102 low-light images validated the feasibility of the replacement by producing PSNR=30.8682(db) and SSIM=0.4345.

• The Korean Journal of Applied Statistics
• /
• v.37 no.3
• /
• pp.311-322
• /
• 2024

In this study, we conduct an exploratory analysis of the MovieLens data, which is frequently used in many recommender system researches, to examine the detailed characteristics of the data. Also, we seek alternatives to improve the well-known neural collaborative filtering (NCF) method. NCF improved matrix factorization method by using deep neural networks in recommender systems. We devise, generalized NCF (G-NCF), a variant of NCF and test the performances. The G-NCF we propose shows superior characteristics on average performance across key evaluation metrics, compared to the NCF, but it also has a slightly larger variance in the evaluation metrics. Evaluation metrics such as MAP and nDCG were considered for comparison.

• Geophysics and Geophysical Exploration
• /
• v.23 no.2
• /
• pp.72-88
• /
• 2020

Recent rapid advances in computer hardware performance have led to relatively low computational costs, increasing the number of applications of machine-learning techniques to geophysical problems. In particular, deep-learning techniques are gaining in popularity as the number of cases successfully solving complex and nonlinear problems has gradually increased. In this paper, applications of seismic data denoising methods using deep-learning techniques are introduced and investigated. Depending on the type of attenuated noise, these studies are grouped into denoising applications of coherent noise, random noise, and the combination of these two types of noise. Then, we investigate the deep-learning techniques used to remove the corresponding noise. Unlike conventional methods used to attenuate seismic noise, deep neural networks, a typical deep-learning technique, learn the characteristics of the noise independently and then automatically optimize the parameters. Therefore, such methods are less sensitive to generalized problems than conventional methods and can reduce labor costs. Several studies have also demonstrated that deep-learning techniques perform well in terms of computational cost and denoising performance. Based on the results of the applications covered in this paper, the pros and cons of the deep-learning techniques used to remove seismic noise are analyzed and discussed.

• The Journal of the Korea Contents Association
• /
• v.21 no.3
• /
• pp.616-625
• /
• 2021

Deep neural networks are an approximation method that approximates an arbitrary function to a linear model and then repeats additional approximation using a nonlinear active function. In this process, the method of evaluating the performance of approximation uses the loss function. Existing in-depth learning methods implement approximation that takes into account loss functions in the linear approximation process, but non-linear approximation phases that use active functions use non-linear transformation that is not related to reduction of loss functions of loss. This study proposes parametric activation functions that introduce scale parameters that can change the scale of activation functions and location parameters that can change the location of activation functions. By introducing parametric activation functions based on scale and location parameters, the performance of nonlinear approximation using activation functions can be improved. The scale and location parameters in each hidden layer can improve the performance of the deep neural network by determining parameters that minimize the loss function value through the learning process using the primary differential coefficient of the loss function for the parameters in the backpropagation. Through MNIST classification problems and XOR problems, parametric activation functions have been found to have superior performance over existing activation functions.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2021.06a
• /
• pp.301-302
• /
• 2021

본 논문에서는 움직임 정보와 시공간 주의 정보를 심층신경망을 이용하여 함께 활용한 행동 인식 방법을 제안한다. RGB 영상을 입력으로 사용하는 기존 방법과 달리 제안하는 방법은 움직임 정보를 입력으로 사용하여 시간적 특징 및 시공간 주의 정보를 추출하고, RGB 영상에서 추출한 공간적 특징에 시공간 주의 정보를 고려하게 하여 행동 인식 정확도를 향상시킨다. 실험 결과를 통해 행동 분류 정확도 및 연산 효율성이 기존 신경망보다 우수함을 보인다.

• Journal of Advanced Navigation Technology
• /
• v.23 no.6
• /
• pp.597-604
• /
• 2019

Artificial neural networks are algorithms that simulate learning through interaction and experience in neurons in the brain and that are a method that can be used to produce accurate results through learning that reflects the characteristics of data. In this study, a model using deep neural network was presented to improve the predicted wind speed values in the meteorological dynamic model. The wind speed prediction improvement model using the deep neural network presented in the study constructed a model to recalibrate the predicted values of the meteorological dynamics model and carried out the verification and testing process and Separate data confirm that the accuracy of the predictions can be increased. In order to improve the prediction of wind speed, an in-depth neural network was established using the predicted values of general weather data such as time, temperature, air pressure, humidity, atmospheric conditions, and wind speed. Some of the data in the entire data were divided into data for checking the adequacy of the model, and the separate accuracy was checked rather than being used for model building and learning to confirm the suitability of the methods presented in the study.