http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
SSVEP 기반 BCI를 위한 RGB 채널로 분할된 시각 자극 설계
심희동(Hee-Dong Sim),양석조(Seok-Jo Yang) 대한기계학회 2020 대한기계학회 춘추학술대회 Vol.2020 No.12
Brain-Computer Interface(BCI) is a technology that provides a direct communication path between the human brain and external devices. Electroencephalography(EEG) is most frequently used in BCI due to advantages such as low-cost, non-invasiveness and high temporal resolution. One of the most well studied is Steady-State Visual Evoked Potential(SSVEP) which is neural activity from the visual cortex that are evoked when the user is focusing flickering visual stimulus. In most SSVEP studies, the visual stimuli are used as square wave signals of a certain frequency, but the range of available frequencies are limited by several factors including monitor’s refresh rate of 60 Hertz. In this study, we propose a new visual stimulus design with three frequency components separated by R, G, B channels. To check the usability of the proposed method, we classified 16-targets SSVEP and resting state from three male subjects using deep neural network, and the classification accuracy was 84.87%. We suggest that proposed method can be used to overcome the limitations of conventional SSVEP stimulus design method.
심희동(Hee-Dong Sim),정미리(Mi-Ri Jeong),이미현(Mi-Hyun Lee),양석조(Seok-Jo Yang) 대한기계학회 2020 大韓機械學會論文集B Vol.44 No.4
대부분의 치매는 비가역적이기 때문에 조기 진단이 매우 중요하다. 하지만 현재 사용되고 있는 치매 진단 방법은 많은 한계점들을 가지고 있다. 이러한 문제를 해결하기 위해 치매와 정상인을 분류하기 위한 다양한 연구가 진행되어 왔다. 본 연구에서는 뇌파를 사용하여 정상인과 치매 환자를 분류하기 위해 신호 처리 방법과 일차원 합성곱 인공신경망 분류기를 제안한다. 치매 환자 13명과 정상인 115명의 뇌파 신호를 사용하였으며, 국소 푸리에 변환된 데이터셋으로 제안된 분류기의 성능을 교차 검증하였다. 제안된 인공신경망의 분류 성능은 두 개의 전극만으로 정확도 86.04%, 민감도 82.53%, 특이도 88.69%의 높은 성능을 나타내었다. 이를 통해 본 연구에서 제안된 일차원 합성곱 인공신경망 분류기와 신호 처리 방법이 정상인과 치매 환자의 뇌파를 높은 정확도로 분류할 수 있음을 확인하였다. Since most dementia cases are irreversible, early diagnosis is critical. However, dementia diagnosis methods have several limitations. In this regard, numerous studies have tried to classify dementia patients and healthy persons. In this study, we propose a 1-dimensional (1D) convolution neural network classifier and signal processing methods to distinguish between dementia patients and healthy persons through electroencephalography (EEG). We used EEG data from 13 dementia patients and 115 healthy people. In addition, a dataset transformed by the short-time Fourier transform was used to cross-validate the proposed artificial neural network classifier. The performance of the proposed classifier was 86.04 % accuracy, 82.53 % sensitivity, and 88.69 % specificity. This is a remarkable performance for only two electrodes. These results confirmed that EEG of healthy persons and dementia patients can be classified accurately with the 1D convolution neural network classifier and the signal processing methods proposed in this study.
근전도 신호와 딥러닝을 이용한 Finger Motion Tracking
진성호(Sung-Ho Jin),심희동(Hee-Dong Sim),안장원(Jang-Won Ahn),최효은(Hyo-Eun Choi),박우진(Woo-Jin Park),지수빈(Soo-Been Ji),양석조(Seok-Jo Yang) 대한기계학회 2022 대한기계학회 춘추학술대회 Vol.2022 No.11
VR controllers are generally used by hand. But they are not suitable for Handicapped with amputated arm joint. Controllors for them are needed because virtual reality technology can become close to our real life. This study aims to implement it through electromyography signal and Deep learning. Data sets are created through digital filtering and time series feature extraction. Using 1d CNN-LSTM, CNN is used to create a feature map of electromyogram signal and LSTM is used to analyze time series associations. This model takes about 370ms to calculate the predicted data. Loss of prediction through train data is 0.00164, and 0.008458 for test data and 0.008688 for validation data.