5.1. Conclusions This study was presented a new method that contribute to enhance efficiency of extract information or diagnose from EEG signal by remove EOG artifacts. Firstly, the definition of EEG, EEG artifacts and EOG artifacts is presented. After that, the next section is about Wavelet Transform as well as Haar Wavelet Transform, a type of Wavelet Transform. The next is Independent component analysis for removing artifacts. The end of the related work chapter is about Deep learning and Sparse Autoencoder. The proposed method to remove EOG artifacts is on the next chapter. Many experimental results and evaluation results indicate that the proposed method effectively remove EOG artifacts (even better than ICA JADE). 5.2. Future Works In the future, the proposed method can be used to in a BCI system that helps eliminate EOG artifacts in an online way. It means the signal after record from device is processed immediately and then display it on the screen. And whats more, expanding to handle more types of artifacts as well as comparing with more methods to improve the proposed method. From there, it is possible to help the EEG signal is cleaner before it is taken for analysis, help doctors to provide better analytical results or even help automatically analysis and evaluation of machines by using EEG signals become more reliable and more accurate.
DEEP WAVELET SPARSE AUTOENCODER TO REMOVE ELECTROOCULOGRAPHY ACKNOWLEDGEMENT Firstly, I would like to express my sincere gratitude to my supervisor Assoc Prof Le Thanh Ha of University of Engineering and Technology, Viet Nam National University, HaNoi for his instructions, guidance and his research experiences Secondly, I am grateful to thank my co-supervisor M.Sc Nguyen The Hoang Anh of Viet Nam Academy of Science and Technology for invaluable assistance during our working time Additionally, I am grateful to thank all the teachers of University of Engineering and Technology, VNU for their invaluable lessons which I have learnt I also thank my friends in K59CA class, University of Engineering and Technology, VNU I greatly appreciate the helps and support from Human Machine Interaction Laboratory of University of Engineering and Technology during this project ABSTRACT The research direction to improve the quality of human life is always a hot issue and highly appreciated The main objective of these studies is to help improve the living environment of people, helping the essential needs and desires of people to be served faster and more accurately without spending too much effort, cost Therefore, based on information about people's wishes, computers can provide appropriate solutions to serve The system using information of the electroencephalogram (EEG) signal may be a solution in this case because in theory the body's wishes, thoughts, and actions are derived from the brain and at a specific time, brain signal is an expression of those desires, thoughts, and actions But extract the information from EEG signal is plagued by many kinds of noise In this work, I will introduce a new method of removing electrooculography (EOG) artifacts, a typical type of noise and difficult to control, in EEG signals This helps improve the ability to extract information from the obtained EEG signal The results of the proposed method are compared with independent components analysis (ICA) method Table of Content List of Figures ix List of Tables xi ABBREVATIONS xii Chapter1 .1 INTRODUCTION 1.1 Motivation 1.2 Contributions and thesis overview .2 1.2.1 Contributions 1.2.2 Thesis structure Chapter2 .4 RELATED WORK 2.1 Electroencephalography (EEG) 2.1.1 Overview .4 2.1.2 EEG applications 2.2 EEG Artifacts and Electrooculography (EOG) Artifacts .6 2.2.1 EEG Artifacts 2.2.1.1 Internal artifacts 2.2.1.2 External artifacts .8 2.2.2 Electrooculography (EOG) Artifacts 2.3 Wavelet Transform and Haar wavelet 2.3.1 Wavelet Transform 2.3.2 Haar wavelet transform 12 2.4 Independent component analysis (ICA) and ICA JADE 12 2.4.1 Independent component analysis (ICA) 12 2.4.2 ICA JADE 15 2.5 Deep learning and Sparse Autoencoder .15 2.5.1 Deep learning 15 2.5.1.1 Overview 15 2.5.1.2 Neural network 16 2.5.2 Sparse Autoencoder 18 2.5.2.1 Overview 18 2.5.2.2 Network architecture 18 Chapte.r3 .20 METHODOLOGY: Deep Wavelet Sparse Autoencoder to remove Electrooculography .20 3.1 Data set and Preprocessing data 20 3.1.1 Data set .20 3.1.2 EOG Detection with Haar Wavelet Transform 20 3.1.3 Preprocessing data 22 3.2 A Deep Wavelet Sparse Autoencoder to remove EOG artifacts 23 3.3 Training and correcting artifacts 25 Chapter4 27 EXPERIMENTS AND RESULTS 27 4.1 Experimental setting 27 4.2 Evaluation metrics .27 4.2.1 Visual assessment 27 4.2.2 Power Spectral Density (PSD) 27 4.2.3 Frequency correlation .28 4.3 Results and discussions 28 Chapte.r5 .37 CONCLUSIONS 37 5.1 Conclusions 37 5.2 Future Works 37 References 38 Appendix 42 List of Figur Figure 2.1 Recorded EEG signal Figure 2.2 Example about eye-related artifacts [14] Figure 2.3 Example about cardiac artifacts [9] Figure 2.4 Example about muscle-related artifacts (Chewing artifact) [10] .7 Figure 2.5 First examples about EOG artifacts Figure 2.6 Second examples about EOG artifacts Figure 2.7 Some basic (mother) wavelets 10 Figure 2.8 Multi-resolution analysis using discrete wavelet transform .11 Figure 2.9 Haar wavelet shape [4] 12 Figure 2.10 The original EEG signal 13 Figure 2.11 The signal after using ICA method 14 Figure 2.12 Example about what components should be chosen to remove (set to 0) 14 Figure 2.13 Example about a neural network with input layer, output layer and hidden layers [25] 17 Figure 2.14 A single neuron 17Y Figure 3.1 Emotiv EPOC+ [13] 20 Figure 3.2 Scalp locations covered by Emotiv EPOC+ [7] .20 Figure 3.3 Main steps of EOG detection using Haar Wavelet Transform 21 Figure 3.4 Example about length of single square-shaped .22 Figure 3.5 Example about single EOG with haar-wavelet segments in length 22 Figure 3.6 Flowchart of algorithm 24 Figure 3.7 Haar Wavelet to decompose original signal at level Figure 4.1 Applying EOG Detection to original EEG signal 28 Figure 4.2 Applying EOG Detection in second EEG signal record 29 Figure 4.3 Applying EOG Detection in third EEG signal record 29 Figure 4.4 The results of DWSAE and ICA JADE compare with original signal29 Figure 4.5 Compare signal corrected by DWSAE and original signal .30 Figure 4.6 Compare signal corrected by ICA JADE and original signal 30 Figure 4.7 Compare signal corrected by DWSAE and original signal in second segment 31 Figure 4.8 Compare signal corrected by ICA JADE and original signal in second segment 31 Figure 4.9 Compare signal corrected by DWSAE and original signal in other record 31 Figure 4.10 Compare signal corrected by ICA JADE and original signal in other record 32 Figure 4.11 Power Spectral Density (PSD) of original signal, signal corrected by ICA JADE and signal corrected by DWSA .34 Figure 4.12 Frequency correlation between original signal and signal correct by ICA JADE 35 Figure 4.13 Frequency correlation between original signal and signal correct by DWSAE 35 List of Tables Table 4.1 The changing of coefficients under 32Hz between before and after using DWSAE 32 ABBREVATIONS EEG Electroencephalogram BCI Brain-computer-interface WT Wavelet Transform EOG Electrooculography ICA Independent component analysis PCA Principal component analysis DWT Discrete Wavelet Transform NN Neural Network DWSAE Deep Wavelet Sparse Autoencoder PSD Power Spectral Density Chapter INTRODUCTION 1.1 Motivation All activities of the body come from the brain and with the desire to be able to understand the brain, understand how the brain gives instructions to the body, researches and studies about brain have been and are still being developed For measuring the electrical activity of the brain, a physiological method was used, Electroencephalogram (EEG) EEG uses small, metal discs (electrodes) attached to the scalp to record electrical activity in the brain The neurons in brain are active all the time and communicate via electrical impulses, even sleeping The electrical activity of the brain shows up as wavy lines on an EEG recording Electroencephalogram (EEG) have been used in various applications, including human–computer interfaces, diagnosis of brain diseases, and measurement of cognitive status For example, doctors use EEG signals obtained from patients for diagnosis and classification then make decisions to improve the health of patients with epilepsy [32] For more, EEG signals can be used in conjunction with BCI systems to control electrical devices such as wheelchairs, which will enable people with disabilities to move on their own, without the need for assistance from others [1] However, while recording the electrical activity of the brain, gathering artifacts is very common and unavoidable Especially eye-related artifacts (EOG artifacts), a type that has the most significant effect and is most difficult to control during EEG collecting process Therefore, artifacts removal is necessary for ensuring the outcome from the analysis and evaluation of the received signal to be not seriously affected Nowadays, there are some methods to eliminate the artifacts in the obtained EEG signal, can be mentioned with names like Independent component analysis (ICA) [19], Principal component analysis (PCA) [3] The ICA and PCA methods both analyze the obtained signal into components Some of these components are identified as artifactual components and then removing the artifacts is done by reconstructing components without artifactual components But the difference is in ICA, components is equally important, while in PCA the components are not It means, the first component of the PCA is the one that best explains the variability of data, the second component is the second best explanation and must be orthogonal to the first one, However, PCA has a big disadvantage that cannot completely separate eye artifacts from brain signals, especially when they have comparable amplitudes ICA algorithms are superior to PCA in removing a wide variety of artifacts from the EEG signal, even in the case of comparable amplitudes but the main disadvantage of ICA is not automatic The component based procedures used for artifact removal have to be chosen manually [18], [23] Therefore, the problem is that find a method to eliminate artifacts in the EEG signal that can work well like ICA method (even better) and can be run automatically Deep Learning may be another approach to solve this problem because Deep Learning could be trained to remove artifacts automatically Since the EEG signal is non-stationary, the frequency is always changing by time, therefore no results can be confirmed to be completely accurate The results are often evaluated through metrics such as Power Spectral Density (PSD), Frequency Correlation 1.2 Contributions and thesis overview 1.1.1 Contributions Types of artifacts, especially Electrooculography (EOG) artifacts, are almost present in Electroencephalogram (EEG) recordings It may lead to a wrong or unpredictable judgment of a doctor when analyzing a patient's EEG signal Or, these artifacts can interfere as well as minimize the possibility of obtaining information automatically from the recorded EEG signal The purpose of this thesis is to propose a method to eliminate automatically EOG artifacts in obtained EEG signals by combining Haar Wavelet Transform and Sparse Autoencoder with multi hidden layers This can lead to applications such as: helping analysis, which are done by people or machines, from EEG signals to be more efficient and accurate; contributing to creating a Brain-computer-interface (BCI) system that remove EOG artifacts in online way (real time) not from EEG pre-recordings 1.1.2 Thesis structure The rest of this thesis is organized as follows Chapter provides background knowledge that is related to Electrooculography (EOG) artifacts removal The main point in this chapter is to supply reader background knowledge about EEG, artifacts and Electrooculography (EOG) artifacts, Wavelet Transform in processing EEG signal, Independent component analysis (ICA) in artifacts removal, Deep Learning as well as Sparse Autoencoder Chapter EXPERIMENTS AND RESULTS 4.1 Experimental setting In my work, the collecting data use Emotiv EPOC+ with BCI system that built before The collected data are saved as ‘.mat’ files (using MATLAB can open it) All steps from preprocessing data, training model to correcting EEG signal are run on Windows 10 (Intel Core i5-8400 CPU2.8GHz (6 CPUs), 16GB RAM The preprocessing step is done by using MATLAB (release R2017b) and EOG Detection method is written on MATLAB Training model and correcting EEG signal are written on Python The weight of trained model and the output of correcting phase are also save as ‘.mat’ files The result of my work is compared with a kind of ICA method, named ICA JADE 4.2 Evaluation metrics To evaluate the results of the proposed method, following metrics are used to evaluate: 4.2.1 Visual assessment Firstly, the most basic evaluation is visualizing the corrected signal by image and evaluating based on the experience of recognizing EOG artifacts (usually evaluated by experts) If in the areas that are corrected, the voltage of signal is not too high or too low for the signal in other areas around which not to be corrected, it is considered a good result, and if there is a sudden change of the signal, the result of method is not good and needs improvement 4.2.2 Power Spectral Density (PSD) The visual assessment method cannot be fully confirmed that the result of the method is good It is necessary to use Power Spectral Density (PSD) as the next measure metric in evaluating the results Power Spectral Density (PSD) is the frequency response of a random or periodic signal It tells where the average power is distributed as a function of frequency Power Spectral Density as a performance metric gives only a rough estimate in providing an inference relating to the relative superiority of algorithms in removing ocular artifacts from EEG In the original EEG signal, there are more EOG artifacts than corrected signal, so the distributed of EOG 27 artifacts frequencies (between 0-16 Hz) in original signal is always more than in corrected signal Therefore, the result of the method is considered acceptable when the distributed of EOG artifacts frequencies is significantly reduced compared to the original signal and the other frequencies are almost no change Additionally, in the corrected signal, the frequencies of EOG artifacts not appear too much compared to other frequencies Conversely, the result is considered bad 4.2.3 Frequency correlation In addition to using PSD for evaluation, there is another method to increase the reliability of the PSD result, named Frequency correlation Frequency correlation calculates the change of each frequency of the corrected signal with the original signal Similar to PSD, the result is rated as good when the change of frequency is mainly focused on the frequencies of EOG artifacts (in the range of 0-16Hz), other regions may change but not allowed too much Otherwise, the result is bad 4.3 Results and discussions From the collected dataset from 12 subjects through 65 experiments, training was made with a length of 26130 128-sample segments, equivalent to nearly 7.26 hours of data collection at sampling rate 128Hz Figure 4.1 describes the results of the applying EOG Detection method to the original signal This image shows that the initial signal has EOG artifacts and the task is to eliminate EOG artifacts in these regions to produce clean signals Figure 4.1 Applying EOG Detection to original EEG signal In other EEG signal record, EOG Detection method also yields quite good results in identifying EOG artifacts, all EOG artifacts are identified Detail in Figure 4.2 and Figure 4.3 28 Figure 4.2 Applying EOG Detection in second EEG signal record Figure 4.3 Applying EOG Detection in third EEG signal record Figure 4.4 The results of DWSAE and ICA JADE compare with original signal 29 Figure 4.4 shows the result of the proposed method (DWSAE) and the ICA JADE method compared to the original data As shown in Figure 4.4, it can be seen that both methods give quite good results in removing EOG artifacts in detected regions But at first glance it was possible to realize that the ICA JADE also affects other areas, not only EOG artifacts while DWSAE does not For greater certainty, see Figures 4.5 and Figure 4.6 Figure 4.5 shows that the DWSAE only changes the signal in the EOG artifacts region without changing the signal in other regions, while Figure 4.6 shows that, in addition to changing the signal in the EOG artifacts region accordingly, the ICA JADE also signal change in other areas too This say that, in visual assessment, DWSAE is better than ICA JADE in eliminating EOG artifacts in EEG signals Some of the resulting images in other segments (even other record) also show similar evaluation results (as shown in Figures 4.7 and Figures 4.8 for other segments, Figure 4.9, Figure 4.10 for other record) Figure 4.5 Compare signal corrected by DWSAE and original signal Figure 4.6 Compare signal corrected by ICA JADE and original signal 30 Figure 4.7 Compare signal corrected by DWSAE and original signal in second segment Figure 4.8 Compare signal corrected by ICA JADE and original signal in second segment Figure 4.9 Compare signal corrected by DWSAE and original signal in other record 31 Figure 4.10 Compare signal corrected by ICA JADE and original signal in other record To explain for this result, the proposed method will reduce the difference between the values of the approximation coefficients (under 16Hz) of the signal to match the normal signal at EOG artifacts positions The appropriate reduction is automatically learned by the network during training The following table describes the change of the approximation coefficients before and after using the DWSAE to eliminate EOG artifacts: Table 4.1 The changing of coefficients under 32Hz between before and after using DWSAE Coefficients Before After Approximate component [147.7615; 69.7615] [60.1603; 44.6481 ] [-28.4999; -80.2499] [47.9876; 58.7960] Detail component level [-78.6656; 103.4143; [67.0964; 46.8257 (5-8Hz) -58.3363; -32.5269] 43.4921; 56.1395] Detail component level [ -3.7500; -58.9999; [54.9949; 58.6262; 79.4999; -5.7500; 37.2751; 44.2898; -2.7499; -21.7499; 67.3138; 62.4343; 6.7499; -1.2499] 54.4460 59.0538 ] [ -1.7677; -2.1213; [ -1.7677; -2.1213; (0-2Hz) Detail component level (3-4Hz) (9-16Hz) Detail component level 32 (17-32Hz) 3.1819; -40.6586; 3.1819; -40.6586; 33.9411; 26.1629; 33.9411; 26.1629; 0.3535; -2.828; 0.3535; -2.8284; 0.7071; 6.7175; 0.7071; 6.7175; -11.6672; 4.9497; -11.6672; 4.9497; -19.0918; -26.5165; -19.0918; -26.5165; 19.7989; -7.4246] 19.7989; -7.4246] As show in Table 4.1, the difference of the values of the approximation coefficients at frequencies between 0-16Hz after using DWSAE is greatly reduced compared to before use Specifically, at 0-2Hz, before using is | 147.7615 - 69.7615 | = 78 and after use is | 60.1603 - 44.6481 | = 15.5122; at the frequency of 3-4Hz, before use is | -28.4999 - (-80.2499) | = 51.75 and after use is | 47.9876 - 58.7960 | = 10.8084; similar in other frequency of 5-16Hz It can also be seen that the coefficients in the frequency range from 17-32Hz are not changed, the reason is that EOG artifacts exists only in the frequency range from 0-16Hz [20] This result is therefore consistent with the characteristics of the EOG artifacts and is considered an important point in evaluating the outcome of the method As mentioned in the evaluation metrics section, if only evaluate with visual assessment, it is not convincing enough Therefore, following the results between the two methods are evaluated by using Power Spectral Density (PSD) Figure 4.11 shows the results of PSD of two methods compared to the original signal 33 Figure 4.11 Power Spectral Density (PSD) of original signal, signal corrected by ICA JADE and signal corrected by DWSA Figure 4.11 shows that the ICA and DWSAE methods both reduce the frequency areas of EOG artifacts (in the range of 0-16Hz [20]) compared to the original signal This is very good but the DWSAE method is seen to be superior to ICA JADE in the following Other frequency regions (> 16Hz) are almost unchanged in the DWSAE method but ICA JADE changes a lot This once again proves that ICA JADE is worse than DWSAE in reducing EOG artifacts To further prove, the last metric of the three metrics mentioned above is used to evaluate the results of the two methods, named the Frequency Correlation Details are shown in Figures 4.12 and 4.13 According to Figures 4.12 and 4.13, one can see that two methods change the distributed of EOG artifacts frequency (0-16 Hz [20]) greatly But these figures also show clearly that other frequency regions (> 16Hz) are still changed in ICA JADE method but not in DWSAE method This, again, confirms that DWSAE is better than ICA JADE in reducing EOG artifacts Figure 4.12 Frequency correlation between original signal and signal correct by ICA JADE 34 Figure 4.13 Frequency correlation between original signal and signal correct by DWSAE All three evaluation metrics above shows that DWSAE eliminates EOG artifacts better than ICA JADE and in addition DWSAE is also better than ICA in time consuming Specifically, the time consuming to eliminate EOG artifacts in a 13168sample signal of ICA JADE is about 15.13 seconds As for DWSAE, the total time of training with 26130x128 samples and correcting the 13168-sample signal is about 16.22 seconds (where more than 16.21 seconds for training and only less than 0.01 second for correcting) And an indispensable thing is the DWSAE method can run automatically from the beginning to the end and ICA JADE cannot run automatically because selecting the components to delete must be done by humans 35 Chapter CONCLUSIONS 5.1 Conclusions This study was presented a new method that contribute to enhance efficiency of extract information or diagnose from EEG signal by remove EOG artifacts Firstly, the definition of EEG, EEG artifacts and EOG artifacts is presented After that, the next section is about Wavelet Transform as well as Haar Wavelet Transform, a type of Wavelet Transform The next is Independent component analysis for removing artifacts The end of the related work chapter is about Deep learning and Sparse Autoencoder The proposed method to remove EOG artifacts is on the next chapter Many experimental results and evaluation results indicate that the proposed method effectively remove EOG artifacts (even better than ICA JADE) 5.2 Future Works In the future, the proposed method can be used to in a BCI system that helps eliminate EOG artifacts in an online way It means the signal after record from device is processed immediately and then display it on the screen And what's more, expanding to handle more types of artifacts as well as comparing with more methods to improve the proposed method From there, it is possible to help the EEG signal is cleaner before it is taken for analysis, help doctors to provide better analytical results or even help automatically analysis and evaluation of machines by using EEG signals become more reliable and more accurate 36 References [1] Abiyev R H., N Akkaya, E Aytac, I Günsel and A Çağman, "Braincomputer interface for control of wheelchair using fuzzy neural networks", BioMed research international 2016, 2016, pp 1-8 [2] Nguyen The Hoang Anh, John Musson, Feng Li, Wei Wang, Guangfan Zhang, Roger Xu, Carl Richey, Tom Schnell, Frederic D McKenzie, Jiang Li, "EOG artifact removal using a wavelet neural network", Neurocomputing 97, 2012, pp 374389 [3] Babu P Ashok, and K V S V R Prasad, "Removal of ocular artifacts from EEG signals using adaptive threshold PCA and wavelet transforms", 2011 International Conference on Communication Systems and Network Technologies IEEE, 2011, pp 572-575 [4] Baker J.W, "Quantitative classification of near-fault ground motions using wavelet analysis", Bulletin of the Seismological Society of America 97.5, 2007, pp 1489 [5] Bell Anthony J., and Terrence J Sejnowski, "An informationmaximization approach to blind separation and blind deconvolution", Neural computation 7.6, 1995, pp 1129-1159 37 [6] Bengio Yoshua, Aaron Courville, and Pascal Vincent, "Representation learning: A review and new perspectives", IEEE transactions on pattern analysis and machine intelligence 35.8, 2013, pp 1798-1828 [7] Botti V., Y Gomez, V Martinez, J Vila, "EEG-based biomarkers and emotional response to the Ultimatum Game", 2014, pp [8] Cardoso Jean-Franỗois, and Antoine Souloumiac, "Blind beamforming for non-Gaussian signals", IEE proceedings F (radar and signal processing), Vol 140 No IET Digital Library, 1993 [9] Cardiac artifacts image [Online] https://www.slideshare.net/SudhakarMarella/eeg-artifacts-15175461 [10] Chewing artifact image [Online] https://www.ncbi.nlm.nih.gov/books/NBK390358/ [11] Cireşan Dan, Ueli Meier, and Jürgen Schmidhuber, "Multi-column deep neural networks for image classification", arXiv preprint arXiv:1202.2745,2012 [12] de Beer, Nicole AM, et al, "Educational simulation of the electroencephalogram (EEG)", Technology and Health Care9.3, 2001, pp 237-256 [13] Emotiv EPOC+ image [Online] https://www.emotiv.com/epoc/ [14] Eye-related artifacts image [Online] https://www.slideshare.net/SudhakarMarella/eeg-artifacts-15175461 [15] Ferreira A., W C Celeste, F A Cheein, T F Bastos-Filho, M Sarcinelli-Filho, R Carelli, "Human-machine interfaces based on EMG and EEG applied to robotic systems", Journal of NeuroEngineering and Rehabilitation 5.1, 2008, pp 10 [16] Hagemann Dirk and Ewald Naumann, "The effects of ocular artifacts on (lateralized) broadband power in the EEG", Clinical Neurophysiology 112.2, 2001, pp 216 [17] Hyvärinen Aapo and Erkki Oja, "Independent component analysis: algorithms and applications", Neural networks 13.4-5, 2000, pp 411-430 [18] Jung T.P., C Humphries, T.W Lee, S Makeig, M.J McKeown, V Iragui, T.J Sejnowski, "Extended ICA removes artifacts from electroencephalographic recordings", Advances in neural information processing systems, 1998, pp 896-897 38 [19] Jung T.P., C Humphries, T.W Lee, S Makeig, M.J McKeown, V Iragui, T.J Sejnowski, "Removing electroencephalographic artifacts: comparison between ICA and PCA", Neural Networks for Signal Processing VIII, Proceedings of the 1998 IEEE Signal Processing Society Workshop (Cat No 98TH8378) IEEE, 1998 [20] Krishnaveni V., S Jayaraman, S Aravind, V Hariharasudhan, K Ramadoss, "Automatic identification and removal of ocular artifacts from EEG using wavelet transform", Measurement science review 6.4, 2006, pp 45, 50 [21] Krishnaveni V., S Jayaraman, L Anitha, K Ramadoss, "Removal of ocular artifacts from EEG using adaptive thresholding of wavelet coefficients", Journal of Neural Engineering 3.4, 2006, pp 338 [22] LeCun, Yann, Yoshua Bengio, learning", nature 521.7553, 2015, pp 436 and Geoffrey Hinton, "Deep [23] Makeig S., A.J Bell, T.P Jung, T.J Sejnowski, "Independent component analysis of electroencephalographic data", Advances in neural information processing systems, 1996, pp 151 [24] Nair, Vinod, and Geoffrey E Hinton, "Rectified linear units improve restricted boltzmann machines", Proceedings of the 27th international conference on machine learning (ICML-10), 2010 [25] Neural network image [Online] https://medium.com/datadriveninvestor/when-not-to-use-neural-networks89fb50622429 [26] Le Van Quyen M., J Martinerie, V Navarro, P Boon, M D'Havé, C Adam, M Baulac, "Anticipation of epileptic seizures from standard EEG recordings", The Lancet 357.9251, 2001, pp 183-188 [27] Rao, T V K H., and Dhongade Dayanand Vishwanath, "Detecting sleep disorders based on EEG signals by using discrete wavelet transform", 2014 International Conference on Green Computing Communication and Electrical Engineering (ICGCCEE), IEEE, 2014 [28] Russell, Jon, "Google's AlphaGo AI wins three-match series against the world's best Go player", TechCrunch https://techcrunch com/2017/05/24/alphagobeats-planets-best-human-go-player-ke-jie/(25 May 2017), 2017 39 [29] Savelainen, Antti, “An introduction to EEG artifacts”, 2010, pp 8-10 [30] Schalk G., D J McFarland, T Hinterberger, N Birbaumer, J R Wolpaw, "BCI2000: a general-purpose brain-computer interface (BCI) system", IEEE Transactions on biomedical engineering 51.6, 2004, pp 1034-1043 [31] Schmidhuber, Jürgen, "Deep learning in neural networks: An overview", Neural networks 61, 2015, pp 85-117 [32] Smith, S J M, "EEG in the diagnosis, classification, and management of patients with epilepsy", Journal of Neurology, Neurosurgery & Psychiatry 76.suppl 2, 2005, pp ii2-ii7 [33] Souloumiac, Antoine, "Nonorthogonal joint diagonalization by combining givens and hyperbolic rotations", IEEE Transactions on Signal Processing 57.6, 2009, pp 2222-2231 [34] Tan D E., R S Tung, W Y Leong, J C M Than, "Sleep disorder detection and identification", Procedia engineering 41, 2012, pp 289-295 [35] Tanaka, Kazuo, Kazuyuki Matsunaga, and Hua O Wang, "Electroencephalogram-based control of an electric wheelchair", IEEE transactions on robotics 21.4, 2005, pp 762-766 [36] Teplan M., “Fundamentals of EEG measurement”, Measurement science review, Vol 2(2), 2002, pp [37] Wilson J A., G Schalk, L M Walton, J C Williams, "Using an EEGbased brain-computer interface for virtual cursor movement with BCI2000", JoVE (Journal of Visualized Experiments) 29, 2009, pp 1319 [38] Wolpaw J R., N Birbaumer, W J Heetderks, D J McFarland, P H Peckham, G Schalk, T M Vaughan, "Brain-computer interface technology: a review of the first international meeting", IEEE transactions on rehabilitation engineering 8.2, 2000, pp 164-173 40 Appendix Published articles: [1] Tran Huy Hoang, Nguyen The Hoang Anh, Do Quoc Vuong and Le Thanh Ha, “EOG Detection using Haar Wavelet Transform toward implementation of an IOT Brain computer interface”, Một số vấn đề chọn lọc Công nghệ thông tin Truyền thông 2018, 2018 41 ... Wavelet Sparse Autoencoder to remove EOG artifacts A Deep Wavelet Sparse Autoencoder (DWSAE) is used to eliminate EOG artifacts from EEG signals This network has almost a structure of Sparse Autoencoder, ... 2.5.2.1 Overview Sparse autoencoder is essentially an autoencoder combined with a sparsity constraint An autoencoder is a type of artificial neural network and learning algorithm of autoencoder is... METHODOLOGY: Deep Wavelet Sparse Autoencoder to remove Electrooculography .20 3.1 Data set and Preprocessing data 20 3.1.1 Data set .20 3.1.2 EOG Detection with Haar Wavelet