• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2019.11a
• /
• pp.102-105
• /
• 2019

본 논문에서는 심층 신경망 검색 방법을 사용하여 이미지 고해상도화를 위한 심층 신경망을 설계하는 방법을 구현하였다. 일반적으로 이미지 고해상도화, 잡음 제거 및 번짐 제거를 위한 심층신경망 구조는 사람이 설계하였다. 최근에는 이미지 분류 등 다른 영상처리 기법에서 사용하는 심층 신경망 구조를 검색하기 위한 방법이 연구되었다. 본 논문에서는 강화학습을 사용하여 이미지 고해상도화를 위한 심층 신경망 구조를 검색하는 방법을 제안하였다. 제안된 방법은 policy gradient 방법의 일종인 REINFORCE 알고리즘을 사용하여 심층 신경망 구조를 출력하여 주는 제어용 RNN(recurrent neural network)을 학습하고, 최종적으로 이미지 고해상도화를 잘 실현할 수 있는 심층 신경망 구조를 검색하여 설계하였다. 제안된 심층 신경망 구조를 사용하여 이미지 고해상도화를 구현하였고, 약 36.54dB 의 피크 신호 대비 잡음 비율(PSNR)을 가지는 것을 확인할 수 있었다.

• Journal of the Korea Computer Graphics Society
• /
• v.27 no.5
• /
• pp.73-79
• /
• 2021

Recently, in the field of computer animation, a method for generating motion using deep learning has been studied away from conventional finite-state machines or graph-based methods. The expressiveness of the network required for learning motions is more influenced by the diversity of motion contained in it than by the simple length of motion to be learned. This study aims to find an efficient network structure when the types of motions to be learned are diverse. In this paper, we train and compare three types of networks: basic fully-connected structure, mixture of experts structure that uses multiple fully-connected layers in parallel, recurrent neural network which is widely used to deal with seq2seq, and transformer structure used for sequence-type data processing in the natural language processing field.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2022.05a
• /
• pp.64-64
• /
• 2022

이상기후로 인해 돌발적이고 국지적인 호우 발생의 빈도가 증가하게 되면서 짧은 선행시간(~3 시간) 범위에서 수치예보보다 높은 정확도를 갖는 초단시간 강우예측자료가 돌발홍수 및 도시홍수의 조기경보를 위해 유용하게 사용되고 있다. 일반적으로 초단시간 강우예측 정보는 레이더를 활용하여 외삽 및 이동벡터 기반의 예측기법으로 산정한다. 최근에는 장기간 레이더 관측자료의 확보와 충분한 컴퓨터 연산자원으로 인해 레이더 자료를 활용한 인공지능 심층학습 기반(RNN(Recurrent Neural Network), CNN(Convolutional Neural Network), Conv-LSTM 등)의 강우예측이 국외에서 확대되고 있고, 국내에서도 ConvLSTM 등을 활용한 연구들이 진행되었다. CNN 심층신경망 기반의 초단기 예측 모델의 경우 대체적으로 외삽기반의 예측성능보다 우수한 경향이 있었으나, 예측시간이 길어질수록 공간 평활화되는 경향이 크게 나타나므로 고강도의 뚜렷한 강수 특징을 예측하기 힘들어 예측정확도를 향상시키는데 중요한 소규모 기상현상을 왜곡하게 된다. 본 연구에서는 이러한 한계를 보완하기 위해 적대적 생성 신경망(Generative Adversarial Network, GAN)을 적용한 초단시간 예측기법을 활용하고자 한다. GAN은 생성모형과 판별모형이라는 두 신경망이 서로간의 적대적인 경쟁을 통해 학습하는 신경망으로, 데이터의 확률분포를 학습하고 학습된 분포에서 샘플을 쉽게 생성할 수 있는 기법이다. 본 연구에서는 2017년부터 2021년까지의 환경부 대형 강우레이더 합성장을 수집하고, 강우발생 사례를 대상으로 학습을 수행하여 신경망을 최적화하고자 한다. 학습된 신경망으로 강우예측을 수행하여, 국내 기상청과 환경부에서 생산한 레이더 초단시간 예측강우와 정량적인 정확도를 비교평가 하고자 한다.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.22 no.2
• /
• pp.127-133
• /
• 2021

A recent study of people physically falling focused on analyzing the motions of the falls using a recurrent neural network (RNN) and a deep learning approach to get good results from detecting 2D human poses from a single color image. In this paper, we investigate a detection method for estimating the position of the head and shoulder keypoints and the acceleration of positional change using the skeletal keypoints information extracted using PoseNet from an image obtained with a low-cost 2D RGB camera, increasing the accuracy of judgments about the falls. In particular, we propose a fall detection method based on the characteristics of post-fall posture in the fall motion-analysis method. A public data set was used to extract human skeletal features, and as a result of an experiment to find a feature extraction method that can achieve high classification accuracy, the proposed method showed a 99.8% success rate in detecting falls more effectively than a conventional, primitive skeletal data-use method.

• Nuclear Engineering and Technology
• /
• v.55 no.3
• /
• pp.814-826
• /
• 2023

Sensor faults in nuclear power plant instrumentation have the potential to spread negative effects from wrong signals that can cause an accident misdiagnosis by plant operators. To detect sensor faults and make accurate accident diagnoses, prior studies have developed a supervised learning-based sensor fault detection model and an accident diagnosis model with faulty sensor isolation. Even though the developed neural network models demonstrated satisfactory performance, their diagnosis performance should be reevaluated considering real-time connection. When operating in real-time, the diagnosis model is expected to indiscriminately accept fault data before receiving delayed fault information transferred from the previous fault detection model. The uncertainty of neural networks can also have a significant impact following the sensor fault features. In the present work, a pilot study was conducted to connect two models and observe actual outcomes from a real-time application with an integrated system. While the initial results showed an overall successful diagnosis, some issues were observed. To recover the diagnosis performance degradations, additive logics were applied to minimize the diagnosis failures that were not observed in the previous validations of the separate models. The results of a case study were then analyzed in terms of the real-time diagnosis outputs that plant operators would actually face in an emergency situation.

• Journal of Internet Computing and Services
• /
• v.24 no.6
• /
• pp.99-105
• /
• 2023

In the 4th industrial revolution, the field of criminal justice is paying attention to Legaltech using artificial intelligence to provide efficient legal services. This paper attempted to create a crime prediction model that can apply Recurrent Neural Network(RNN) to increase the potential for using legal technology in the domestic criminal justice field. To this end, the crime process was divided into pre, during, and post stages based on the criminal facts described in the judgment, utilizing crime script analysis techniques. In addition, at each time point, the method and evidence of crime were classified into objects, actions, and environments based on the sentence composition elements and the 8 principles of investigation. The case summary analysis framework derived from this study can contribute to establishing situational crime prevention strategies because it is easy to identify typical patterns of specific crime methods. Furthermore, the results of this study can be used as a useful reference for research on generating crime situation prediction data based on RNN models in future follow-up studies.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.19 no.11
• /
• pp.310-318
• /
• 2018

Recently, the importance of big data analysis is increasing as a large amount of data is generated by various devices connected to the Internet with the advent of Internet of Things (IoT). Especially, it is necessary to analyze various large-scale IoT streaming sensor data generated in real time and provide various services through new meaningful prediction. This paper proposes a real-time indoor PM10 concentration prediction LSTM model based on streaming data generated from IoT sensor using AWS. We also construct a real-time indoor PM10 concentration prediction service based on the proposed model. Data used in the paper is streaming data collected from the PM10 IoT sensor for 24 hours. This time series data is converted into sequence data consisting of 30 consecutive values from time series data for use as input data of LSTM. The LSTM model is learned through a sliding window process of moving to the immediately adjacent dataset. In order to improve the performance of the model, incremental learning method is applied to the streaming data collected every 24 hours. The linear regression and recurrent neural networks (RNN) models are compared to evaluate the performance of LSTM model. Experimental results show that the proposed LSTM prediction model has 700% improvement over linear regression and 140% improvement over RNN model for its performance level.

• Journal of the Korean Society of Systems Engineering
• /
• v.18 no.2
• /
• pp.94-107
• /
• 2022

Accidents prevention and mitigation is the highest priority of nuclear power plant (NPP) operation, particularly in the aftermath of the Fukushima Daiichi accident, which has reignited public anxieties and skepticism regarding nuclear energy usage. To deal with accident scenarios more effectively, operators must have ample and precise information about key safety parameters as well as their future trajectories. This work investigates the potential of machine learning in forecasting NPP response in real-time to provide an additional validation method and help reduce human error, especially in accident situations where operators are under a lot of stress. First, a base-case SGTR simulation is carried out by the best-estimate code RELAP5/MOD3.4 to confirm the validity of the model against results reported in the APR1400 Design Control Document (DCD). Then, uncertainty quantification is performed by coupling RELAP5/MOD3.4 and the statistical tool DAKOTA to generate a large enough dataset for the construction and training of neural-based machine learning (ML) models, namely LSTM, GRU, and hybrid CNN-LSTM. Finally, the accuracy and reliability of these models in forecasting system response are tested by their performance on fresh data. To facilitate and oversee the process of developing the ML models, a Systems Engineering (SE) methodology is used to ensure that the work is consistently in line with the originating mission statement and that the findings obtained at each subsequent phase are valid.

• Journal of Broadcast Engineering
• /
• v.24 no.3
• /
• pp.495-505
• /
• 2019

Typical algorithms of deep learning include CNN(Convolutional Neural Networks), which are mainly used for image recognition, and RNN(Recurrent Neural Networks), which are used mainly for speech recognition and natural language processing. Among them, CNN is able to learn from filters that generate feature maps with algorithms that automatically learn features from data, making it mainstream with excellent performance in image recognition. Since then, various algorithms such as R-CNN and others have appeared in object detection to improve performance of CNN, and algorithms such as YOLO(You Only Look Once) and SSD(Single Shot Multi-box Detector) have been proposed recently. However, since these deep learning-based detection algorithms determine the success of the detection in the still images, stable object tracking and detection in the video requires separate tracking capabilities. Therefore, this paper proposes a method of combining Kalman filters into deep learning-based detection networks for improved object tracking and detection performance in the video. The detection network used YOLO v2, which is capable of real-time processing, and the proposed method resulted in 7.7% IoU performance improvement over the existing YOLO v2 network and 20 fps processing speed in FHD images.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.40 no.5
• /
• pp.449-456
• /
• 2022

This study proposes a deep learning algorithm to predict crop yield using GIS (Geographic Information System) to extract soil properties from Soilgrids and soil suitability class maps. The proposed model modified the structure of a published CNN-RNN (Convolutional Neural Network-Recurrent Neural Network) based crop yield prediction model suitable for the domestic crop environment. The existing model has two characteristics. The first is that it replaces the original yield with the average yield of the year, and the second is that it trains the data of the predicted year. The new model uses the original field value to ensure accuracy, and the network structure has been improved so that it can train only with data prior to the year to be predicted. The proposed model predicted the yield per unit area of autumn cabbage for kimchi by region based on weather, soil, soil suitability classes, and yield data from 1980 to 2020. As a result of computing and predicting data for each of the four years from 2018 to 2021, the error amount for the test data set was about 10%, enabling accurate yield prediction, especially in regions with a large proportion of total yield. In addition, both the proposed model and the existing model show that the error gradually decreases as the number of years of training data increases, resulting in improved general-purpose performance as the number of training data increases.