Journal of the Korea Academia-Industrial cooperation Society Vol. 19, No. 10 pp. 523-529, 2018 https://doi.org/10.5762/kais.2018.19.10.523 ISSN 1975-4701 / eissn 2288-4688 박종준, 권춘기 * 순천향대학교의료 IT 공학과 A Consistency Study of CNN's Learning to Recognize Korean Finger Number using semg Signals Jong-Jun Park, Chun-Ki Kwon * Department of Medical IT Engineering, Soonchunhyang University 요약합성곱신경망 (Convolutional Neural Network, CNN) 은컴퓨터비전분야에서활발히적용되어왔으며, 이미지분류, 문서분류, 지문인식등에서탁월한인식능력을보여왔음을여러연구를통해서검증되었다. 본연구는시계열의표면근전도신호를입력데이터로취하는숫자지화인식응용에이미지분류에서탁월한인식성능을보이는합성곱신경망을적용한것으로, 반복적인한국숫자지화인식수행에서도일관된학습을수행하는지를검증하는연구로, 문헌에서보기힘든연구이다. 이를검증하기위해, 한국숫자지화영 (0) 부터다섯 (5) 까지의여섯숫자지화를시연하도록훈련한실험대상 1 인의아래팔근육으로부터획득한숫자별 60 개씩총 360 개의표면근전도신호를획득하였으며, 그중에서 252 개의표면근전도신호를입력데이터, 108 개의표면근전도신호는테스트데이터로 CNN 인식에활용하였다. CNN 인식을위해필요한학습단계는 100 학습단계, CNN 인식의반복수행횟수는 10 회로설정하였으며, 반복수행마다테스트데이터를활용하여인식률을계산하였다. 본연구에서실험한결과에서보듯이, 반복인식마다 CNN 의학습은일관되었으며, 99.1% 이상 (60 숫자지화중하나의숫자지화인식에오류발생 ) 의높은인식률을보였다. 따라서, CNN 기법은시계열의표면근전도신호를입력데이터로하는숫자지화인식분야에서도전역솔루션과함께우수한인식능력을제공하는기법중에하나이다. Abstract Convolutional Neural Network (CNN) has been actively employed in the application of computer vision, and has been proved to have its superior performance in image classification, document classification, and finger print recognition. This work focuses on an application of CNN, having outstanding performance in image classification, to recognition of korean finger number using time series semg signals as input and validates CNN's capability in providing its consistent learning in repeated application for recognition of semg based Korean finger numbers, which has been rarely a topic in previous studies. To this end, 252 semg signals as input data and 108 semg signals as test data out of 360 semg signals (60 signals each number) acquired from a forearm muscle of the subject who is trained to consistently perform six Korean finger number gestures from zero(0) to five(5) were used for CNN based finger number recognition. CNN was set to have 100 learning iterations for each application of finger number recognition, and to have 10 repetitive applications of finger number recognition for the consistency of CNN's learning. Recognition rate at each repetition was calculated from test data. As can be seen from the results in this work, CNN shows consistent learning at each repetitive application of finger number recognition and outstanding recognition rates of more than 99.1% (missed one case out of 60 cases). Thus, CNN is one of powerful techniques for finger number recognition based on time-series semg signals to provide not only global solution but also excellent recognition rates. Keywords : convolutional neural network, time-series signal, surface electromyography, Korean finger number gesture recognition, consistency in cnn learning, repeated recognition application 본논문은일부순천향대학교의지원을받아수행되었음. * Corresponding Author : Chun-Ki Kwon(Soonchunhyang Univ.) Tel: +82-41-530-3091 email: chunkikwon@sch.ac.kr Received July 25, 2018 Revised (1st August 13, 2018, 2nd August 20, 2018) Accepted October 5, 2018 Published October 31, 2018 523
한국산학기술학회논문지제 19 권제 10 호, 2018 1. 서론 합성곱신경망 (Convolutional Neural Network, CNN) 은컴퓨터비전분야에서활발히적용되어, 이미지분류, 문서분류, 지문인식등에서탁월한인식능력을여러문헌들에서검증되었다 [1-6]. 기존의분류방법은입력데이터의특징을다양한특징추출법으로추출하고, 이를 Linear Support Vector Machine (LSVM) 등의분별알고리즘을적용하여대상을분류한다 [7-9]. 이와는달리, 합성곱신경망은학습을통해입력데이터의특징추출과대상분류를동시에수행한다 [1-6]. 이러한 CNN 의우수한인식능력은시계열데이터인음성신호를분류하는음성인식분야에까지입증되고있다 [1]. 한편, 표면근전도 (Surface Electromyograph, semg) 신호의응용은초기에는단순히근육의활성도유무를판별하여 On/Off의스위치기능으로많이사용되어왔으나, 표면근전도신호처리와알고리즘의등장으로분별이가능한카테고리가다양해짐에따라휠체어의방향제어는물론숫자지화를인식하는분야에까지확장되었다 [10-19]. 최근의표면근전도신호를활용한숫자지화인식연구는숫자영 (0) 부터아홉 (9) 까지의중국숫자지화를아래팔근육에부착된네 (4) 채널의전극을통해측정된표면근전도신호를활용하였다 [17-18]. 이들연구는기존의영상인식방법과마찬가지로숫자시연시의발생되는표면근전도신호의특징을특징추출법으로추출하고분별알고리즘을통해숫자지화를인식하는것으로높은인식률을보이고있다. 하지만, 숫자지화시연시의표면근전도신호의특징을추출하고분류하는특징추출법과분류알고리즘을연구자의경험에의존하는시행착오적인방법으로선정하는등의한계가있다. 본연구는특징추출과특징분류를학습을통해자동적으로처리하는 CNN 기법을시계열데이터인표면근전도신호를활용한한국숫자지화인식에적용한초기연구로서, CNN이표면근전도신호를입력데이터로반복적인인식응용에서도일관된학습을수행하는지를검증한다. 이를위해, Fig. 1에서보여지는숫자지화영 (0) 부터다섯 (5) 의여섯 (6) 한국숫자지화를선정하고, 여섯 (6) 한국숫자지화시연시에획득한표면근전도신호를입력데이터로처리하여, 10회의반복적인인식응용수행에서 CNN이일관된학습을수행하였음을확인하였다. Fig. 1. Korean finger numbers gestures of number zero to five selected 따라서, 표면근전도신호에기반한한국숫자지화인식응용에서 CNN은일관된학습수행을통하여전역솔루션과우수한인식능력을제공하기때문에표면근전도신호기반한국숫자지화인식응용에적용할수있는우수한기법이다. 또한, 본연구는 CNN이입력데이터가이미지데이터에만국한되지않고시계열인표면근전도신호를입력데이터로활용할수있음을보였으며, 이는시계열신호인표면근전도신호를입력데이터로숫자지화인식에 CNN을적용한첫연구이다. 2. 본론 2.1 한국숫자지화인식을위한표면근전도신호측정 2.1.1 표면근전도신호측정시스템본연구에서표면근전도신호를측정하기위해사용된근전도계측시스템은네덜란드의바이오세미 (Biosemi) 사에서연구목적용으로개발한 Active-II 시스템이다. Fig. 2에서보는바와같이, 바이오세미사의 Active-II 계측시스템은능동전극, 아날로그-디지털변환박스, 배터리박스로구성된전방부와변환된디지털신호값을 PC로전달하는데이터수신기와배터리충전기, 그리고계측소프트웨어를구동하기위한컴퓨터로구성된후방부로나누어진다 [20]. Fig. 2. Configuration of Active-II System to acquisition semg signals 524
2.1.2 실험대상본연구의목적은동일한표면근전도신호를입력데이터로활용한한국숫자지화인식에적용된 CNN이수행하는학습의일관성을검증하는초기연구로서, 숫자지화별로균일한표면근전도신호의확보가필요하다. 따라서, 전극위치, 손가락움직임성향등으로실험대상별로발생하는영향을최소화하기위하여한국숫자지화시연에어려움이없는건전한남성 1명을선정하였다. 선정된대상은숫자지화별로균일한표면근전도신호가나타나도록한국숫자지화시연에대한숙련도를높이도록훈련을하였다. 본연구에서는선정된한국숫자지화별로 60개씩총 360개의샘플을획득하였으며, 이중에서 252개의샘플을학습데이터, 108개의샘플을테스트데이터로활용하였다. 지화인다섯 (5) 의시연에서활성된다. 손가락폄근 (extensor digitorum) : 아래팔뒤 가쪽근에위치하며, 집게손가락, 가운뎃손가락, 약손가락, 그리고새끼손가락의펴기에관여하는주요근육으로숫자지화하나 (1) 에서넷 (4) 까지의숫자지화시연시에근전도신호가활성화된다. 한국숫자지화와선정된네개의아래팔근육과전극의위치를 Table. 1에정리하였으며, 선정된아래팔근육들의활성도를측정하기위해부착된전극의위치를실연시의모습으로 Fig. 3에도시하였다 [19]. Table 1. Main forearm muscles for flexor and extensor associated with Korean Finger Number Gestures 2.1.3 표면근전도신호측정을위한아래팔근육과전극위치표면근전도신호를활용한숫자영 (0) 부터다섯 (5) 까지의한국숫자지화인식을위해손가락움직임과관련되는아래팔근육의선정과전극의위치는매우중요하다 [15-17]. 최근의연구인표면근전도신호를활용한숫자영 (0) 부터아홉 (9) 까지의중국숫자지화연구에서는네가지의아래팔근육으로도우수한숫자지화인식률을보이고있다 [17]. 본연구에서도 Fig. 1에서의여섯한국숫자지화시연시에표면근전도신호의분별이가능하도록얕은손가락굽힘근 (flexor digitorum superficialis), 긴엄지굽힘근 (flexor policis longus), 긴엄지벌림근 (abductor prolicis longus), 그리고손가락폄근 (extensor digitorum) 의네개의아래팔근육을선정하였다 [19,21]. Muscles Selected Electrode Channel flexor digitorum superficialis flexor policis longus abductor policis longus extensor digitorum CH1 CH2 CH3 CH4 얕은손가락굽힘근 (flexor digitorum superficialis) : 아래팔앞 안쪽근에위치하며집게손가락부터새끼손가락까지의굽힘에관여하는주요근육으로숫자영 (0) 부터셋 (3) 의시연시에활성화가된다. 긴엄지굽힘근 (flexor policis longus) : 아래팔앞 안쪽근에위치하며, 엄지손가락의굽힘에관여하는주요근육중하나이며엄지손가락의굽힘이필요한숫자지화에서근전도활성도를보인다. 긴엄지벌림근 (abductor prolicis longus) : 아래팔뒤 가쪽근에위치하며, 엄지손가락의벌림에관여하는근육중하나이며, 벌리는동작을하는숫자 Fig. 3. Electrical placement for four muscles selected (a) CH1 and CH2 (b) CH3 and CH4 2.1.4 한국숫자지화별표면근전도신호본연구에서는측정된표면근전도신호는 10-250Hz 의 대역폭을가지는 2차버터워스대역필터 (2 nd -Butterworth Filter) 를적용하였고, 센서등의오프셋값등에따른잡음을최소화하기위하여기준선이동 (Baseline Shift) 을통하여신호의기준값을 0으로셋팅하였다. 숫자영 (0) 525
한국산학기술학회논문지제 19 권제 10 호, 2018 부터다섯 (5) 까지의한국숫자지화인식을위해선정된아래팔근육과해당전극위치로부터측정한표면근전도신호의샘플을 Fig. 4에도시하였으며, 각숫자지화시연에따른표면근전도신호가숫자지화별로특징이나타나는것을볼수있다 [19]. 하여학습한 CNN의한국숫자지화영 (0) 부터다섯 (5) 에대하여 108개의테스트데이터를활용하여계산한 CNN의인식률결과를 Fig. 5에도시하였다. Fig. 5에서보는바와같이, 학습이이루어지지않은초기에는인식률이현저히떨어지지만학습단계가거듭될수록증가하여학습단계가 80회를넘어가면서인식률이 100% 로수렴됨을볼수있다. Table 2는 100 학습단계후의컨퓨전매트릭스 (Confusion Matrix) 를보여준것으로서, 왼쪽의실제시연된숫자지화와위쪽의 CNN 이인식하여예측한숫자가 100% 로일치하여, 대각선으로만값이나타나는것으로볼수있다. Fig. 4. semg signals collected from each electrode pair for Korean finger number gestures from 0 to 5 2.2 표면근전도신호를활용한한국숫자지화인식에의 CNN 적용 2.2.1 CNN 기법적용을위한학습환경표면근전도신호를활용하여한국숫자지화를인식하기위한 CNN (Convolutionary Neural Network) 학습은오픈소스딥러닝라이브러리인텐서플로우 (Tensorflow) 를사용하여수행하였으며, 빠른실행을위해행렬이나벡터계산에특화된그래픽처리장치 (Graphic Processing Unit, GPU) 를활용하였다 [22]. 본연구에서 NVIDIA사의 GTX-970 에탑재된 4GB 메모리를가지는 GPU를사용하였다. 측정된한국숫자지화별로 60개씩총 360개의표면근전도신호중에서한국숫자지화별로 42개씩 252개의샘플을학습데이터, 18개씩 108개의샘플을테스트데이터로활용하여 CNN 학습을수행하였다. 2.2.2 CNN 기법을활용한한국숫자지화인식결과표면근전도신호 252개의샘플을학습데이터로활용 Fig. 5. Validation of CNN based Korean finger number gesture recognition at learning iteration Table 2. CNN s confusion matrix after 100 th learning iteration Finger Number Data Actual Finger Number 0 18 Predicted Finger Number 0 1 2 3 4 5 1 18 2 18 3 18 4 18 5 18 Accurary (%) 100 100 100 100 100 100 2.3 한국숫자지화인식에서의 CNN 학습의일관성검증 2.3.1 한국숫자지화인식에적용된 CNN 학습의일관성검증을위한환경설정앞서기술한 2.2절에서보여주었듯이, 여섯 (6) 개의한국숫자지화에대하여취득한표면근전도신호를 CNN 526
에적용하여 100 학습단계로학습하여인식하는것을인식수행하나 (1) 의셋이라고할때, 인식률이 100% 임을보임으로서, 그유용성이충분히입증되었다. 본절에서는표면근전도신호를입력데이터로활용하는 CNN의학습이반복적인인식수행에서도일관된학습을보이는지를검증하고자한다. CNN 학습의일관성을보이기위해, 동일한학습데이터에대하여총 10번의반복적인인식수행을하였으며, 반복적인인식수행셋에서 CNN의학습변화는테스트데이터로확인하였다. 2.3.2 한국숫자지화인식에적용된 CNN 의일관성검증 Fig. 6은반복적인인식수행셋마다 CNN이학습하는변화를보여주고있다. Fig. 6에서보는바와같이 CNN 의학습하는경향이모든인식수행셋들에서비슷하게나타나고있다. 즉, CNN에의한학습은학습이거의되어있지않은초기단계에서는학습변화가크지만, 학습이어느정도수준에이를수록학습속도가느려짐을보여주고있다. 또한, 100단계의학습을한후에는여섯 (6) 개의숫자에대한숫자지화를거의 100% 로인식하고있어인식률이반복적인인식수행에서도매우높아학습이제대로되었음을보여주고있다. 수준이지만, 여섯숫자전체로에대한인식률로확대하면, 총 108번의숫자지화인식에대하여 1회의오류에해당하는 99.1% 로거의 100% 에가까운인식률이다. 또한, 표 4에서보는바와같이, 일곱번째인식수행을제외한여타인식수행들에서는여섯숫자지화모두각각제대로인식하여모두 100% 의인식률을보여주고있다. 이러한결과가보여주듯이, 표면근전도신호를활용한한국숫자지화인식에있어서 CNN의학습이매우일관되게이루어지는것을검증하였다. 본연구에서시계열데이터인표면근전도신호를활용한 CNN 기반숫자지화인식은조명등의주변환경에영향을받는기존의영상인식방법에서의제약을극복함은물론, 문헌 [17] 과 [18] 에서수행한중국숫자지화인식연구에서처럼시행착오적으로선정되는기존의특징추출법과분류알고리즘을병행하여활용하는숫자지화인식률인 95% 보다훨씬뛰어난 99.1% 이상의인식률을보인다. 또한기존의특징추출법과분류알고리즘의선정방법은국소솔루션의가능성을배제할수없지만, CNN의일관된학습경향은전역솔루션을제공하고있음을보여주고있다. Table 3. Confusion matrix after 100 th learning iteration at 7 th repetition Finger Number Data Actual Finger Number Predicted Finger Number 0 1 2 3 4 5 0 17 1 remark 1 18 2 18 3 18 4 18 5 18 Accuracy (%) 94.1 100 100 100 100 100 99.1 Fig. 6. Change of the recognition rate at every learning iteration during 100 iterations of CNN at each recognition process for CNN based Korea finger number recognition CNN이 100 단계학습을수행하는반복적인인식수행셋마다각각의숫자지화별로인식률을보여주는컨퓨전매트릭스를표3와표4와같이살펴보았다. 표 3에서보는바와같이, 일곱번째인식수행에서실제숫자영 (0) 을시행한 18번의숫자지화인식에대하여 1회를숫자다섯 (5) 으로인식하여인식률이 94.1% 로다소높은 Table 4. Confusion matrix after 100 th learning iteration at all of repetitions except 7 th repetition Finger Number Data Actual Finger Number 0 18 Predicted Finger Number 0 1 2 3 4 5 remark 1 18 2 18 3 18 4 18 5 18 Accuracy (%) 100 100 100 100 100 100 100 527
한국산학기술학회논문지제 19 권제 10 호, 2018 3. 결론 본연구에서는영상인식분야에서탁월한우수성이입증한 CNN을시계열데이터인표면근전도신호를활용한한국숫자지화인식에적용하여, 반복적인인식학습시에도일관된학습을수행하는지를검증하였다. 한국숫자지화영 (0) 부터다섯 (5) 의여섯한국숫자지화시연시에획득한표면근전도신호의입력데이터에대하여수행한 10회의반복적인인식수행에서 CNN은모두 100% 의인식률을보이는일관된학습경향을보였다. 따라서, CNN은국소적인해결책을찾는학습을수행하는것이아니라, 전역적인해결책을찾는학습을수행함을검증하였다. 따라서, CNN은표면근전도신호를활용한한국숫자지화인식에우수한접근방법중의하나임을알수있다. 향후에는본연구의실용화를위해한국숫자지화의개수확대는물론다수의실험대상자로부터표면근전도신호의균일성확보방안등의추가적인연구가필요하다. References [1] O. Abdel-Hamid, A. R. Mohamed, H. Jiang, L. Deng, G. Penn, D. Yu, Convolutional Neural Networks for speech recognition, IEEE/ACM Transactions on Audio, Speech, and Language Processing, Vol.22, No.10, pp.1533-1545, 2014. DOI: https://dx.doi.org/10.1109/taslp.2014.2339736 [2] W. Shang, K. Sohn, D. Almeida, H. Lee, Understanding and Improving Convolutional Neural Networks via Concatenated Rectified Linear units, Proceedings of the 33 rd International Conference on Machine Learning, New York, USA, 2016, arxiv:1603.05201v2 [3] C. Zhang, K. Qiao, L. Wang, L. Tong, Y. Zeng, B. Yan, Constraint-Free Natural Image Reconstruction From fmri Signals Based on Convolutional Neural Network, Frontiers in Human Neruoscience, Vol.12, 2018. DOI: https://dx.doi.org/10.3389/fnhum.2018.00242 [4] A. Mahendran, A. Vedaldi, Visualizing Deep Convolutional Neural Networks Using Natural Pre-images, International Journal of Computer Vision, Vol.120, No.3, pp.233-255, 2016. DOI: http://dx.doi.org/10.1007/s11263-016-0911-8 [5] C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelow, D. Erhan, V. Vanhoucke, A. Rabinovich, Going deeper with convolutions, Proceedings of Computer Vision and Pattern Recognition, pp.1-9, 2015. DOI: https://dx.doi.org/10.1109/cvpr.2015.7298594 [6] P. Li, F. Zhao, Y. Li, Z. Zhu, Law text classification using semi-supervised convolutional neural networks, Proceedings of 2018 Chinese Control And Decision Conference, pp.309-313, 2018. DOI: https://dx.doi.org/10.1109/ccdc.2018.8407150 [7] V. S. Kulkarni, S. D. Lokhande, Appearance Based Recognition of American Sign Language Using Gesture Segmentation, International Journal on Computer Science and Engineering, Vol.2, No.3, pp.560-565, 2010. [8] H. D. Yang, S. H. Lee, Automatic Extraction of Sign Language and Finger Gestures for Continuous Recognition, Journal of Electrical Engineering and Information Science: Software and Its application, Vol.38, No.2, pp.102-107, 2011. [9] N. H. Kim, A Development of the Next-generation Interface System Based on the Finger Gesture Recognizing in Use of Image Process Techniques, Journal of the Korea Institute of Information and Communication Engineering, Vol.15, No.4, pp.935-942, 2011. DOI: https://dx.doi.org/10.6109/jkiice.2011.15.4.935 [10] H.S. Kang, Finger Sign Recognition Technique using semg Sensor and Gyro Sensor, Master Thesis, Soongsil University, 2002. [11] J. M. Hahne, B. Graimann, K. R. Muller, Spatial Filtering for Robust Myoelectric Control, IEEE Transactions on Biomedical Engineering, Vol.59, No.5, pp.1436-1443, 2012. DOI: https://dx.doi.org/10.1109/tbme.2012.2188799 [12] A. Phinyomark, P. Phukpattaranont, C. Limsakul, A Review of Control Methods for Electric Power Wheelchairs Based on Electromyography Signals with Special Emphasis on Pattern Recognition, IETE Technical Review, Vol.28, No.4, pp.316-326, 2011. DOI: https://dx.doi.org/10.4103/0256-4602.83552 [13] P. J. Lin, H. Y. Chen, Design and implement of a rehabilitation system with surface electromyography technology, Proceedings of 2018 IEEE International Conference on Applied System Invention (ICASI), pp.513-515, 2018. DOI: https://dx.doi.org/10.1109/icasi.2018.8394300 [14] O. S. Powar, K. Chemmangat, Feature selection for myoelectric pattern recognition using two channel surface electromyography signals, Proceedings of TENCON 2017-2017 IEEE Region 10 Conference, pp.1022-1026, 2017. DOI: https://dx.doi.org/10.1109/tencon.2017.8228007 [15] A. J. Young, L. J. Hargrove, T. A. Kuiken, Improving Myoelectric Pattern Recognition Robustness to Electrode Shift by Changing Interelectrode Distance and Electrode Configuration, IEEE Transactions on Biomedical Engineering, Vol.59, No.3, pp.645-652, 2012. DOI: https://dx.doi.org/10.1109/tbme.2011.2177662 [16] L. Pan, D. Zhang, N. Jiang, X. Sheng, X. Zhu, Improving robustness against electrode shift of high density EMG for myoelectric control through common spatial patterns, Journal of NeuroEngineering and Rehabilitation, Vol.12, No.110, pp.1-16, 2015. DOI: https://dx.doi.org/10.1186/s12984-015-0102-9 [17] X. Xun, Z. J. Wang, Pattern recognition of number gestures based on a wireless surface EMG system, Biomedical Signal Processing and Control, Vol.8, No.2, pp.184-192, 2013. DOI: https://dx.doi.org/10.1016/j.bspc.2012.08.005 528
[18] X. Zhang, X. Chen, Y. Li, V. Lantz, K. Wang, J. Yang, A Framework for Hand Gesture Recognition Based on Accelerometer and EMG Sensors, IEEE Transactions on Systems, Man, and Cybernetics - Part A: Systems and Humans, Vol.41, No.6, pp.1064-1076, 2011. DOI: https://dx.doi.org/10.1109/tsmca.2011.2116004 [19] J. J. Park, Study on semg Signal based Finger Sign Recognition using Convolutional Neural Network, Master Thesis, Soonchunhyang University, 2017. [20] Active Two User Manual Version 3.2, Biosemi, 2007 [21] Faculty Association of Korea Anatomy and Physiology, Human Anatomy, Hyunmoonsa, pp.214-218, 2009. [22] M. Abadi, A. Agarwal, P. Barham, E. Brevdo, Z. Chen, C. Citro, G. S. Corrado, A. Davis, J. Dean, M. Devin, S. Ghemawat, I. Goodfellow, A. Harp, G. Irving, M. Isard, Y. Jia, R. Jozefowicz, L. Kaiser, M. Kudlur, J. Levenberg, D. Mané, R. Monga, S. Moore, D. Murray, C. Olah, M. Schuster, J. Shlens, B. Steiner, I. Sutskever, K. Talwar, P. Tucker, V. Vanhoucke, V. Vasudevan, F. Viégas, O. Vinyals, P. Warden, M. Wattenberg, M. Wicke, Y. Yu, X. Zheng, Tensorflow: Large-Scale Machine Learning on Heterogeneous Distributed Systems, in CORR, Vol.abs/160.04467, March 2016. 박종준 (Jong-Jun Park) [ 정회원 ] 2017 년 8 월 : 순천향대학교순천향대학원의료 IT 공학과 ( 공학석사 ) < 관심분야 > 생체신호처리, CNN 기반영상인식 권춘기 (Chun-Ki Kwon) [ 정회원 ] 1994 년 2 월 : 고려대학교고려대학원전기공학과 ( 공학석사 ) 2005 년 8 월 : 퍼듀대학교퍼듀대학원전기전산공학과 ( 전기공학박사 ) 2003 년 3 월 ~ 2006 년 2 월 : 한국연구소책임연구원 2008 년 3 월 ~ 현재 : 순천향대학교의료 IT 공학과교수 < 관심분야 > 생체신호처리, CNN 기반영상인식 529