• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.11 no.6
• /
• pp.3143-3165
• /
• 2017

We present a computerized framework for producing color pastel painting from the visual information extracted from a photograph. To express color pastel painting, we propose a multi-layered framework where each layer possesses pastel stroke patterns of different colors. The stroke patterns in the separate layers are merged by a rendering equation based on a participating media rendering scheme. To produce the stroke patterns in each layer, we review the physical properties of pastels and the mechanism of a convolution framework, which is the most widely used scheme to simulate stick-shaped media such as pencils. We devise the following computational models to extend the convolution framework to produce pastel strokes: a bold noise model, which mimics heavy and clustered deposition of pigment, and a thick convolution filter model, which produces various pastel stroke patterns. We also design a stochastic color coordination scheme to mimic pastel artists' color expression and to separate strokes in different layers. To demonstrate the soundness of approach, we conduct several experiments using the models and compare the results with existing works or real pastel paintings. We present the results for several pastel paintings to demonstrate the excellent performance of our framework.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2018.05a
• /
• pp.703-704
• /
• 2018

An artificial neural network design that traces the pupil for the disables suffering from Lou Gehrig disease is introduced. It grasps the position of the pupil required for the communication system. Tensorflow is used for generating and learning the neural network, and the pupil position is determined through the learned neural network. Convolution neural network(CNN) which consists of 2 stages of convolution layer and 2 layers of complete connection layer is implemented for the system.

• Journal of Information Processing Systems
• /
• v.19 no.5
• /
• pp.688-701
• /
• 2023

The conventional methods of network intrusion detection system (NIDS) cannot measure the trend of intrusiondetection targets effectively, which lead to low detection accuracy. In this study, a NIDS method which based on a deep neural network in a big-data environment is proposed. Firstly, the entire framework of the NIDS model is constructed in two stages. Feature reduction and anomaly probability output are used at the core of the two stages. Subsequently, a convolutional neural network, which encompasses a down sampling layer and a characteristic extractor consist of a convolution layer, the correlation of inputs is realized by introducing bidirectional long short-term memory. Finally, after the convolution layer, a pooling layer is added to sample the required features according to different sampling rules, which promotes the overall performance of the NIDS model. The proposed NIDS method and three other methods are compared, and it is broken down under the conditions of the two databases through simulation experiments. The results demonstrate that the proposed model is superior to the other three methods of NIDS in two databases, in terms of precision, accuracy, F1- score, and recall, which are 91.64%, 93.35%, 92.25%, and 91.87%, respectively. The proposed algorithm is significant for improving the accuracy of NIDS.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.21 no.1
• /
• pp.144-150
• /
• 2017

Recently, deep learning is used for intelligent processing and accuracy improvement of data. It is formed calculation model composed of multi data processing layer that train the data representation through an abstraction of the various levels. A category of deep learning, convolution neural network is utilized in various research fields, which are human pose estimation, face recognition, image classification, speech recognition. When using the deep layer and lots of class, CNN that show a good performance on image classification obtain higher classification rate but occur the overfitting problem, when using a few data. So, we design the training network based on convolution neural network and trained our image data set for object classification in few class problem. The experiment show the higher classification rate of 7.06% in average than the previous networks designed to classify the object in 1000 class problem.

• The Journal of Korean Institute of Electromagnetic Engineering and Science
• /
• v.15 no.10 s.89
• /
• pp.977-982
• /
• 2004

In this paper, a dispersive anisotropic perfectly matched layer(APML) is proposed using piecewise linear recursive convolution(PLRC) for finite difference time domain(FDTD) methods. This proposed APML can be utilized for the analysis of a nonlinear dispersive medium as absorbing boundary condition(ABC). The formulation is simple modification to the original AMPL and can be easily implemented. Also it has some advantages of the PLRC approach-fast speed, low memory cost, and easy formulation of multiple pole susceptibility. We applied this APML to 2-D propagation problems in dispersive media such as Debye and Lorentz media The results showed good absorption at boundaries.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2019.07a
• /
• pp.281-284
• /
• 2019

본 논문은 CNN (Convolution Neural Networks)의 첫 번째 컨볼루션층(convolution layer)을 RGB-csb(RGB channel separation block)로 대체하여 입력 영상의 RGB 값을 특징 맵에 적용시켜 정확성을 제고시킬 수 있는 선행연구 결과에 추가적으로, 훈련 및 시험 영상 수에 따른 분석을 통하여 정확도 향상방법을 제안한다. 제안한 방법은 영상의 개수가 작을수록 각 학습 간의 정확도 편차가 크게 나타나는 불안정성은 있지만 기존 CNN모델에 비하여 정확도 차이가 증가함을 알 수 있다.

• Journal of Information Processing Systems
• /
• v.19 no.1
• /
• pp.130-138
• /
• 2023

The traffic flow in an urban area is affected by the date, weather, and regional traffic flow. The existing methods are weak to model the dynamic road network features, which results in inadequate long-term prediction performance. To solve the problems regarding insufficient capacity for dynamic modeling of road network structures and insufficient mining of dynamic spatio-temporal features. In this study, we propose a novel traffic flow prediction framework called shared spatio-temporal attention convolution optimization network (SSTACON). The shared spatio-temporal attention convolution layer shares a spatio-temporal attention structure, that is designed to extract dynamic spatio-temporal features from historical traffic conditions. Subsequently, the graph optimization module is used to model the dynamic road network structure. The experimental evaluation conducted on two datasets shows that the proposed method outperforms state-of-the-art methods at all time intervals.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2021.06a
• /
• pp.444-444
• /
• 2021

하천의 오염부하량 관리 계획은 지속적인 모니터링을 통한 자료 구축과 모형을 이용한 예측결과를 기반으로 수립된다. 하천의 모니터링과 예측 분석은 많은 예산과 인력 등이 필요하나, 정부의 담당 공무원 수는 극히 부족한 상황이 일반적이다. 이에 정부는 전문가에게 관련 용역을 의뢰하지만, 한국과 같이 지형이 복잡한 지역에서의 오염부하량 배출 특성은 각각 다르게 나타나기 때문에 많은 예산 소모가 발생 된다. 이를 개선하고자, 본 연구는 합성곱 신경망 (convolution neural network)과 수문학적 이미지 (hydrological image)를 이용하여 강우 발생시 BOD 및 총인의 부하량 예측 모형을 개발하였다. 합성곱 신경망의 입력자료는 일반적으로 RGB (red, green, bule) 사진을 이용하는데, 이를 그래도 오염부하량 예측에 활용하는 것은 경험적 모형의 전제(독립변수와 종속변수의 관계)를 무너뜨리는 결과를 초래할 수 있다. 이에, 본 연구에서는 오염부하량이 수문학적 조건과 토지이용 등의 변수에 의해 결정된다는 인과관계를 만족시키고자 수문학적 속성이 내재된 수문학적 이미지를 합성곱 신경망의 훈련자료로 사용하였다. 수문학적 이미지는 임의의 유역에 대해 2차원 공간에서 무차원의 수문학적 속성을 갖는 grid의 집합으로 정의되는데, 여기서 각 grid의 수문학적 속성은 SCS 토양보존국(soil conservation service, SCS)에서 발표한 수문학적 토양피복형수 (curve number, CN)를 이용하여 산출한다. 합성곱 신경망의 구조는 2개의 Convolution Layer와 1개의 Pulling Layer가 5회 반복하는 구조로 설정하고, 1개의 Flatten Layer, 3개의 Dense Layer, 1개의 Batch Normalization Layer를 배열하고, 마지막으로 1개의 Dense Layer가 연결되는 구조로 설계하였다. 이와 함께, 각 층의 활성화 함수는 정규화 선형함수 (ReLu)로, 마지막 Dense Layer의 활성화 함수는 연속변수가 도출될 수 있도록 회귀모형에서 자주 사용되는 Linear 함수로 설정하였다. 연구의 대상지역은 경기도 가평군 조종천 유역으로 선정하였고, 연구기간은 2010년 1월 1일부터 2019년 12월 31일까지로, 2010년부터 2016년까지의 자료는 모형의 학습에, 2017년부터 2019년까지의 자료는 모형의 성능평가에 활용하였다. 모형의 예측 성능은 모형효율계수 (NSE), 평균제곱근오차(RMSE) 및 평균절대백분율오차(MAPE)를 이용하여 평가하였다. 그 결과, BOD 부하량에 대한 NSE는 0.9, RMSE는 1031.1 kg/day, MAPE는 11.5%로 나타났으며, 총인 부하량에 대한 NSE는 0.9, RMSE는 53.6 kg/day, MAPE는 17.9%로 나타나 본 연구의 모형은 우수(good)한 것으로 판단하였다. 이에, 본 연구의 모형은 일반 ANN 모형을 이용한 선행연구와는 달리 2차원 공간정보를 반영하여 오염부하량 모의가 가능했으며, 제한적인 입력자료를 이용하여 간편한 모델링이 가능하다는 장점을 나타냈다. 이를 통해 정부의 물관리 정책을 위한 의사결정 및 부족한 물관리 분야의 행정력에 도움이 될 것으로 생각된다.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.25 no.9
• /
• pp.1158-1165
• /
• 2021

This paper proposes a CNN accelerator which is optimized Pooling layer operation incorporated in Multiplication And Accumulation(MAC) to reduce the memory size. For optimizing memory and data path circuit, the quantized 8bit integer weights are used instead of 32bit floating-point weights for pre-training of MNIST data set. To reduce chip area, the proposed CNN model is reduced by a convolutional layer, a 4*4 Max Pooling, and two fully connected layers. And all the operations use specific MAC with approximation adders and multipliers. 94% of internal memory size reduction is achieved by simultaneously performing the convolution and the pooling operation in the proposed architecture. The proposed accelerator chip is designed by using TSMC65nmGP CMOS process. That has about half size of our previous paper, 0.8*0.9 = 0.72mm². The presented CNN accelerator chip achieves 94% accuracy and 77us inference time per an MNIST image.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.22 no.5
• /
• pp.185-191
• /
• 2022

In this study, we propose a novel super-resolution neural network based on U-Net, residual neural network, and sub-pixel convolution. To prevent the loss of detailed information due to the max pooling of U-Net, we propose down-sampling and connection using sub-pixel convolution. This uses all pixels in the filter, unlike the max pooling that creates a new feature map with only the max value in the filter. As a 2×2 size filter passes, it creates a feature map consisting only of pixels in the upper left, upper right, lower left, and lower right. This makes it half the size and quadruple the number of feature maps. And we propose two methods to reduce the computation. The first uses sub-pixel convolution, which has no computation, and has better performance, instead of up-convolution. The second uses a layer that adds two feature maps instead of the connection layer of the U-Net. Experiments with a banchmark dataset show better PSNR values on all scale and benchmark datasets except for set5 data on scale 2, and well represent local area patterns.