• Journal of IKEEE
• /
• v.24 no.3
• /
• pp.777-782
• /
• 2020

In this paper, we propose CNN-based HMI system using Doppler radar and voice sensor, and present hardware design and implementation results. To overcome the limitation of single sensor monitoring, the proposed HMI system combines data from two sensors to improve performance. The proposed system exhibits improved performance by 3.5% and 12% compared to a single radar and voice sensor-based classifier in noisy environment. In addition, hardware to accelerate the complex computational unit of CNN is implemented and verified on the FPGA test system. As a result of performance evaluation, the proposed HMI acceleration platform can be processed with 95% reduction in computation time compared to a single software-based design.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2022.07a
• /
• pp.81-84
• /
• 2022

제조 산업에서의 이상치 검출은 생산품의 품질과 운영비용을 절감하기 위한 중요한 요소로 최근 딥러닝을 사용하여 자동화되고 있다. 이상치 검출을 위한 딥러닝 기법에는 CNN이 있으며, CNN을 계층적으로 구성할 경우 단일 CNN 모델에 비해 상대적으로 성능의 향상을 보일 수 있다는 것이 많은 선행 연구에서 나타났다. 이에 MVTec-AD 데이터셋을 이용하여 계층 CNN이 다중 클래스 이상치 판별 문제에 대해 효과적인지를 탐구하고자 하였다. 실험 결과 단일 CNN의 정확도는 0.7715, 계층 CNN의 정확도는 0.7838로 다중 클래스 이상치 판별 문제에 있어 계층 CNN 방식 접근이 다중 클래스 이상치 탐지 문제에서 알고리즘의 성능을 향상할 수 있음을 확인할 수 있었다. 계층 CNN은 모델과 파라미터의 개수와 리소스의 사용이 단일 CNN에 비하여 기하급수적으로 증가한다는 단점이 존재한다. 이에 계층 CNN의 장점을 유지하며 사용 리소스를 절약하고자 하였고 K-means, GMM, 계층적 클러스터링 알고리즘을 통해 제작한 새로운 클래스를 이용해 계층 CNN을 구성하여 각각 정확도 0.7930, 0.7891, 0.7936의 결과를 얻을 수 있었다. 이를 통해 Clustering 알고리즘을 사용하여 적절히 물체를 분류할 경우 물체에 따른 개별 상태 판단 모델을 제작하는 것과 비슷하거나 더 좋은 성능을 내며 리소스 사용을 줄일 수 있음을 확인할 수 있었다.

• KIPS Transactions on Software and Data Engineering
• /
• v.11 no.6
• /
• pp.245-254
• /
• 2022

Proteins are the basic unit of all life activities, and understanding them is essential for studying life phenomena. Since the emergence of the machine learning methodology using artificial neural networks, many researchers have tried to predict the function of proteins using only protein sequences. Many combinations of deep learning models have been reported to academia, but the methods are different and there is no formal methodology, and they are tailored to different data, so there has never been a direct comparative analysis of which algorithms are more suitable for handling protein data. In this paper, the single model performance of each algorithm was compared and evaluated based on accuracy and speed by applying the same data to CNN, LSTM, and GRU models, which are the most frequently used representative algorithms in the convergence research field of predicting protein functions, and the final evaluation scale is presented as Micro Precision, Recall, and F1-score. The combined models CNN-LSTM and CNN-GRU models also were evaluated in the same way. Through this study, it was confirmed that the performance of LSTM as a single model is good in simple classification problems, overlapping CNN was suitable as a single model in complex classification problems, and the CNN-LSTM was relatively better as a combination model.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.51 no.2
• /
• pp.69-75
• /
• 2014

Existing People Counting System using a single overhead mounted camera has limitation in object recognition and counting in various environments. Those limitations are attributable to overlapping, occlusion and external factors, such as over-sized belongings and dramatic light change. Thus, this paper proposes the new concept of People Counting System by Facial Age Group using two depth cameras, at overhead and frontal viewpoints, in order to improve object recognition accuracy and robust people counting to external factors. The proposed system is counting the pedestrians by five process such as overhead image processing, frontal image processing, identical object recognition, facial age group classification and in-coming/out-going counting. The proposed system developed by C++, OpenCV and Kinect SDK, and it target group of 40 people(10 people by each age group) was setup for People Counting and Facial Age Group classification performance evaluation. The experimental results indicated approximately 98% accuracy in People Counting and 74.23% accuracy in the Facial Age Group classification.

• Korean Journal of Optics and Photonics
• /
• v.22 no.2
• /
• pp.108-113
• /
• 2011

In this paper, we propose the complex differential diagram to classify surface plasmon waveguide modes with single interface. To date, surface plasmon waveguide modes are classified using the sign change of the group velocity in the dispersion relation that describes the interrelations between the real wavenumber of the propagation direction and the photon energy. The surface plasmon waveguide modes have the wavenumbers of the direction perpendicular to that in which the wave propagates as well as of the propagation direction, so it is necessary to classify the modes using all of these wavenumbers. The complex differential diagram is a graphical representation with variables of the difference between the real component and the imaginary component of the wavenumber. Using this diagram, the specific mode classification is possible, and it is easy to comprehensively analyze the wavenumber as the function of the photon energy.

• Journal of Digital Policy
• /
• v.2 no.3
• /
• pp.1-7
• /
• 2023

In this paper, I proposes a method for performing single image super resolution by separating texture-spatial domains and then classifying features based on detailed information. In CNN (Convolutional Neural Network) based super resolution, the complex procedures and generation of redundant feature information in feature estimation process for enhancing details can lead to quality degradation in super resolution. The proposed method reduced procedural complexity and minimizes generation of redundant feature information by splitting input image into two channels: texture and spatial. In texture channel, a feature refinement process with step-wise skip connections is applied for detail restoration, while in spatial channel, a method is introduced to preserve the structural features of the image. Experimental results using proposed method demonstrate improved performance in terms of PSNR and SSIM evaluations compared to existing super resolution methods, confirmed the enhancement in quality.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2023.01a
• /
• pp.39-42
• /
• 2023

최근 온라인 게임 산업이 급속도로 확장됨과 더불어 Gamebot과 같은 비정상적인 프로그램으로 인한 게임 서비스 피해사례가 급격하게 증가하고 있다. 특히, 대표적인 게임 장르 중 하나인 FPS(First-Person Shooter)에서 Aimbot의 사용은 정상적인 이용자들에게 재미 요소를 잃어버리게 하고 상대적 박탈감을 일으켜 게임의 수명을 줄이는 원인이 된다. 비정상 게임 이용자의 근절을 위해서 메모리 변조 및 불법 변조 프로그램 접근 차단 기법과 불법 프로그램 사용의 패턴 모니터링과 같은 기법들이 제안되었지만, 우회 프로그램 및 새로운 패턴을 이용한 비정상적인 프로그램의 개발에는 취약하다는 단점이 있다. 따라서, 본 논문에서는 정상적인 게임 이용자의 패턴만 학습함으로써 비정상 이용자 검출을 가능하게 하는 딥러닝 기반 단일 클래스 분류 기법을 제안하며, 가장 빈번하게 발생하는 치트(Cheat) 유형인 FPS 게임 내 Aimbot 사용 감지에 초점을 두었다. 제안된 비정상 게임 이용자 감지 시스템은 정상적인 사용자의 마우스 좌표를 데카르트 좌표계(Cartesian coordinates)와 극좌표계(Polar coordinates)의 형태로 패턴을 추출하는 과정과 정상적인 마우스 동작 기록으로 부터 학습된 LSTM 기반 Autoencoder의 복원 에러에 따른 검출 과정으로 구성된다. 실험에서 제안된 모델은 FPS 게임 내 마우스 동작을 기록한 공개 데이터셋인 CSGO 게임 데이터셋으로 부터 학습되었으며, 학습된 모델의 테스트 결과는 데카르트 좌표계로부터 훈련된 제안 모델이 비정상 게임 이용자를 분류하는데 적합함을 입증하였다.

• Journal of Dental Rehabilitation and Applied Science
• /
• v.26 no.1
• /
• pp.21-32
• /
• 2010

The convergence angle of a prepared tooth is a very important factor in the retention and resistance of a crown restoration. But various intraoral environments and clinician's techniques make it difficult to obtain the ideal inclination. Therefore, in this study, clinical convergence angle of a prepared tooth was investigated. The data was collected from the patient models of prosthodontic residents and the patient models of general practitioners. The images of mesiodistal and buccolingual surfaces were taken with a digital camera to evaluate the convergence angle on 'ImageJ' program. The images were classified according to the criteria (1. Clinician group, 2. Position in the dental arch, 3. The purpose of abutment preparation)and then analyzed. The mean convergence angle of a prepared tooth for Korean clinicians was $15.02^{\circ}$ (${\pm}10.13^{\circ}$). 1. It was significant in the convergence angle between the general practitioner group and the prosthodontic resident group(p<0.05). 2. It was significant between the mesiodistal and buccolingual surface in the the prosthodontic resident group(p<0.05). 3. For the general practitioner group, it was significant when anteriors and premolars were compared with molars(p<0.05). For the prosthodontic resident group, it was significant when anteriors and premolars were compared with molars (p<0.05). 4. When divided into upper and lower arches, for the general practitioner group, it showed significant difference in the buccolingual aspect(p<0.05). Also in the prosthodontic resident group, it showed significant difference in the buccolingual aspect(p<0.05). 5. Dividing left and right sides of the arches, there was no significant difference in the general practitioner group and the prosthodontic resident group(p>0.05). 6. In the general practitioner group, it was significant in the mesiodistal axial convergence angle of single crown abutment and 3 unit bridge abutment(p<0.05). In the prosthodontic resident group, it was significant in the mesiodistal and overall axial convergence angle of single crown abutment and 3 unit bridge abutment(p<0.05). Clinical convergence angle of prepared tooth in Korea was included in agreement with other studies investigating convergence angle that ranged from 10 to 22 degrees, achieved in clinical practice.

• The Journal of the Acoustical Society of Korea
• /
• v.39 no.6
• /
• pp.592-599
• /
• 2020

This paper proposes a Convolutional Recurrent Neural Net (CRNN) structure that can simultaneously reflect both static and dynamic characteristics of seismic waveforms for various earthquake events classification. Addressing various earthquake events, including not only micro-earthquakes and artificial-earthquakes but also macro-earthquakes, requires both effective feature extraction and a classifier that can discriminate seismic waveform under noisy environment. First, we extract the static characteristics of seismic waveform through an attention-based convolution layer. Then, the extracted feature-map is sequentially injected as input to a multi-input single-output Long Short-Term Memory (LSTM) network structure to extract the dynamic characteristic for various seismic event classifications. Subsequently, we perform earthquake events classification through two fully connected layers and softmax function. Representative experimental results using domestic and foreign earthquake database show that the proposed model provides an effective structure for various earthquake events classification.

• Journal of the Korean Geographical Society
• /
• v.38 no.3
• /
• pp.375-388
• /
• 2003

This study mainly focuses on evaluating how the triangular training operator could improve classification accuracy in SAR(Synthetic Aperture Radar) and VIR FCC(Visible Infra-red, False Colour Composite). The techniques for the determination of the most informative SAR/VIR combinations in the triangular space diagram, as developed tv the author of the paper, are given and the results obtained are presented. The SAR alone, VIR alone and SAR/VIR FCC classification showed trends for gradual improvement of accuracy. Accuracy distribution pattern for individual classes could be explained closely related to SAR/VIR signature components in the process of the triangular synergistic training. Due to contribution of SAR signature in training samples, it was possible to isolate major terrain features such as cloud cover area and roughness target with acceptable spatial precision. It is anticipated that this research output could be used as a valuable reference for distribution trends of classification accuracy obtained by triangular channel space based training in synergistic application.