Performance Analysis of EEG Signal Processing Based Device Control Applications
Turan Goktug ALTUNDOGAN*, Mehmet Karakose+, Ayse Beyza Gokbulut+, Ilhan Aydin+
*Computer Engineering, Manisa Celal Bayar University, Manisa, Turkey, + Computer Engineering, Firat University, Elazig, Turkey,
turan.altundogan@cbu.edu.tr, mkarakose@firat.edu.tr, gokbulutb@gmail.com, iaydin@firat.edu.tr Abstract— Nowadays, many types of devices are controlled by
electroselenography (EEG) signals. In the literature and in daily life, related studies with EEG controlled devices are increasing day by day. EEG based control applications are applied on many devices such as robot arm, robot, vehicle and unmanned aerial vehicle (UAV). EEG based control procedures usually involve taking, pre-processing, classifying EEG signals, and applying the resulting command to the controlled device. In this study, a performance analysis was carried out by examining the control application studies using EEG signals in the literature. In this analysis study, firstly all studies related to the subject in the literature are examined and the devices, methods, signal processing techniques and classification algorithms used in these studies are handled separately. Appropriate electrode selection for the type of device used in device control applications using EEG signals and type of interaction for command extraction from EEG signal appears to be an important step. In this respect, performance correlations between the types of EEG devices used in the literature studies and the electrode choices used in these studies were compared. Since there are a variety of preprocessing steps for EEG signals, this study provides comparisons based on EEG signal preprocessing techniques. Artificial neural networks (ANN), support vector machines (SVM) and K nearest neighbors (Knn) are used to classify the works in the literature. In this study, comparative studies based on classification methods used in literature studies are also included. As a result, in this study, the studies in the literature for the device control using the EEG signal are examined, compared, interpreted and evaluated, and the points to be considered in the designs to be performed in this area are given. Keywords— EEG Signal, Control, Brain-Computer Interaction
I. INTRODUCTION
Nowadays, many device control processes based on the use of Electro-Electrolenography (EEG) signals can be performed. EEG signals are the generic name given to signals captured by EEG devices as a result of human brain activities. EEG devices were first used in the field of health to control the brain activities of neurological patients. The signals generated by the EEG devices after on output enable the use of certain pre-processing and classification algorithms in the determination of brain activities in device control procedures. The device control process with EEG signals begins by first identifying and activating the electrode according to the brain activity that will perform the control operation. These received signals are then subject to certain preprocessing algorithms. After the preliminary step is passed to the classification step with online or offline learning methods. Once the classification step is over, the control of the relevant device is performed according to the scenario of problem. There are many studies in the literature focused on classification of EEG signals and device control [1-29]. When
the studies in the literature are examined, the preprocessing algorithms used in the device control process can vary. In the literature, in the device control process, band-pass filters are used, depending on the frequency range of the brain activity to be used for user command, and other pre-coding techniques used for noise cleaning. In addition, methods such as artificial neural networks (ANN), support vector machines (SVM), adaptive neural fuzzy inference systems (ANFIS), linear discriminant analysis (LDA) and K nearest neighbors algorithm (Knn) were used for classification. In this study, studies focused on EEG signal based device control are examined in the literature and detailed information about these studies is given. The literature studies examined were compared according to the EEG device type, preprocessing algorithms, classification algorithm, device control method and classification performance. Also in this study, the systems that are developed for device control based on EEG signals have been given attention to the points that should be considered in the design.
II. WORKS FOCUSEDON DEVICE CONTROL USING EEG SIGNALSINTHE LITERATURE
As mentioned earlier, there are many studies in the literature focused on device control using EEG signals. Some of them have been focused on robot and robot arm control using EEG signals [2-18]. In a study, Bahattachary et al. developed an algorithm for multiple classifications based on type 2 fuzzy logic for brain-computer-based control of robot arm [1]. The choice of classifier is very important because the EEG signals that provide the brain computer communication are not stationary, and it is an open issue to investigate. The researchers have developed a type-2 fuzzy system to improve the uncertainties, believing that the adaptive neural output system (ANFIS) will cause ambiguities in the EEG signals. Thus, a proposal for a solution based on a multi-class selective algorithm, an intermittent type-2 fuzzy system and an ANFIS combination is presented in the study. In the developed system, the Jaco robot arm developed by Kinova was controlled by generating motor signals depending on the states of forward, backward, left, right and no movement. EEG signals. The mental states of the users were recorded using an Emotiv Epoc System. This system consists of a wireless neural head with two references and fourteen sensors (electrodes). The EEG system has a sampling frequency of 128 Hz and a resolution of 0.51 μV. Eleven (five male, six female) subjects participated in the betting using the right hand without any discomfort to the study. In order to avoid any external noise, the data reception was carried out in an empty room, protected against noise. Once the data retrieval
has been performed, the data processing step is started. The data processing step contains EEG signal acquisition, preprocessing, property extraction and classification steps. The block diagram of the data processing step is as shown in Fig 1.
Fig. 1 . Block diagram which belongs to study [2].
The proposed method completed training and control over 11 subjects with a performance of 88.91% and 90.11% in the training phase.
In another study, Alomari H. M. and his team performed right and left hand motion classification operations by performing an enhanced feature extraction based on machine learning from EEG signals [3]. In the mentioned study, it is aimed to find out the feature extraction which can best be able to find the differences between the classification algorithms and the punch movements of the right and left hands. The set of EEG data used in the work done, BCI 2000 Inst. obtained from the PyshioNet device manufactured by the company. The data obtained were pre-processed on the Matlab EEGLAB toolbox. The dataset used contained more than 1500 EEG signal bits. The proposed method first takes and filters EEG signals. Filtering is performed with a bandpass filter between 0.5Hz and 90Hz. MATLAB's EEGLAB toolbox is used for the filtering step. The reason for the filtering step is that it is known that EEG signals are quite noisy signals. Following the filtration step, automatic artificiality removal (AAR) is passed. Such an order is needed because EEG signals are very complex and have large physiological useless artificially. Artifacts with AAR step are automatically removed from the EEG data based on blind source separation method. Stages are determined according to the specific activity types of the EEG data that continue after the AAR process. After the AAR process and the phase determination process, the independent component analysis (ICA) step is started. The ICA step was used to separate the electrocortical sources affected by the subjects from the EEG signals. The final preliminary step on the EEG data is the Rhythm Isolation step. This step is completed by applying a bandpass filter with a frequency between 8 and 30 Hz. After analysis of the EEG data set, the activation vector of the gene is calculated. Then the average, power and energy activation are calculated to obtain the feature vector. Here, 6 feature vectors have been obtained and they are expressed by a matrix of power (8 features), energy (8 features), type (1 feature) and chapter (1 feature). Artificial Neural Networks and Support
Vector Machines were used as machine learning methods for classification in the study. In all experiments, 80% of the data was used for training and the remaining 20% was used for the test. This step was repeated 10 times and the data were mixed every time. The best accuracy recorded with the SVM classifier was 89.8% and the highest accuracy recorded with ANN was 97.1%.
I
n another literature study, a mobile robot control process based on Brain Computer Interaction (BCI) was performed [4]. The work is to develop an approach based on a brain signal for people with disabilities who do not drive wheelchairs on their own. After the preprocessing step, property deductions in the time and frequency domain of the signal have been performed. In the feature extraction step of the proposed method, the discrete wave transformation method is used (DWT). After the feature extraction step, linear discriminant analysis (LDA), support vector machines (SVM) and artificial neural networks (ANN) were used to classify the signals. In the class, a total of 60 signal numbers are trained from the dataset and the remaining 20 signals are used for testing purposes. The trained signal contains 15 signals of each test. The classified classification distinguishes the movements of the brain signals from forward, backward, right and left. Forward classification accuracy was 100%, 90% for the left, 80% for the left, and 90% for the right. The accuracy of the command was 95% for the forward classification, 90% for the reverse, 80% for the left, and 85% for the right.Fig. 2 . Block diagram which belongs to study [3].
In addition, other studies in the literature have performed vehicle control procedures using EEG signals [19-29].
In another study, simulation experiments under the control of an EEG vehicle based on imaginary movements were emphasized [20]. The fourteen channel brain electroencephalography (EEG) signal transmitted by the Emotional Epoc + Neuro Headset was collected. Instead of emotiv's own processing and analysis methods, in the developed program, the researchers used their Joint Spatial Pattern (CSP) to extract their wCCA pre-processing algorithms and EEG feature for the pre-processing of EEG signals. Then, the properties obtained by the CNN algorithm are imaginatively classified into the right-handed movement and neutral state of the legs corresponding to three control commands (rotation, right turn, stop) respectively to control the motions of the simulated car. In sigviewer2.2.0. In addition, the accuracy of categorization that applies online learning, time-saving and labor-intensive offline training, and which is closer to the real-time situation of the subjects, has been applied to a large extent. The study is the basis for a practical application for the EEG controlled car. 14 channel EEG signals were used in the study. The frequency of the EEG signals is 128 Hz. The obtained signals were subjected to wavelet-enhanced canonical correlation analysis (wCCA) for noise and artifact removal. The main purpose of using this method is to minimize the loss of cerebral information while finishing artifacts in a very good way.
In the feature extraction step, the common spatial pattern (CSP) method is applied. After the feature extraction step, the signals are classified by the KNN algorithm. The performance of the classification was 84.47% on average.
In another study, a control structure with an electroencephalography (EEG) based BCI system for independent four-wheel electric vehicles was proposed [21]. The BCI system first obtains raw EEG data from the Emotiv EPOC EEG, uses the Independent Component Analysis (ICA) to preprocess the engine image EEG, then uses the Common Spatial Pattern (CSP) to extract the most relevant features it uses. Finally, while the Back Propagation Neural Network (BPNN) is used to classify the left and right directions, it is also calculated to classify the mean threshold value, acceleration and brake of the main band. In the test, the approximate accuracy value for classifying right and left was 84% and the approximate accuracy value for classifying 84% gas and brake was 89.92%. It was later developed for a real-model electric vehicle, including the ultrasonic wave radars and cameras, which is a peripheral perception subsystem that can take the risk off when the BCI command is false. The block diagram of the enhanced system is as shown in Fig 3.
Fig. 2 Block diagram of the proposed method given in the literature study [21].
In another study, a new approach is proposed for the interpretation of electroencephalogram (EEG) signals of drivers to determine the intent of braking for brain-controlled vehicles [23]. Regularization linear discriminant analysis with spatial frequency properties is applied to establish the detection model. These spatial frequency properties are selected from the powers of the frequency points in the sixteen channels using the sequential forward sliding search. Experimental results obtained from twelve testees show that the proposed method can detect intentional braking intentions 420 ms after the onset of emergencies and the feasibility of developing a practical system for detecting the intent of the pilot emergency braking system with over 94% system accuracy. The EEG device used is based on the international 10-20 system. In the developed system, the pre-processing applied to the EEG signals is a bandpass filter with a range of 0.53 to 60 Hz. The feature extraction used is based on alpha, beta, teta and gamma waves. The developed system has been tested on a virtual environment.
Fig. 4 Block diagram of the proposed method given in the literature study [23].
III. GENERAL ALGORITHMIC STEPSOF LITERATURE WORKS As mentioned earlier, the methods given in studies focusing on device control based on EEG signal processing generally consist of four main steps. These consist of taking the EEG data from the user, signal preprocessing, feature extraction and classification. The general block diagram of EEG based device control applications is shown in Fig 5.
Fig. 5. General block diagram of EEG Signal Processing Based Device Control works.
It is useful to examine the steps mentioned here separately. There are different methods that each step uses to perform its operations. Referencing the theory of betting methods will be useful before transmitting comparative results.
A. DATA STEPANDFEATURE EXTRACTION
When the applications in the literature are examined, the control process begins with the acquisition of EEG signals for the realization of the brain computer interface. There are various devices in the market that perform EEG signal reception. These are vary according to the some properties like number of and sequence of electrode, the mode of communication , etc. One of the most important operations of the signal reception step is to determine which electrodes of the EEG device should be used when data is received. EEG
devices used in literature studies are generally in the international standard 10-20 system. Bandpass filter, multi-fractal detrend wave analysis (MFDA) and Fast Fourier Transform (FFT) are used in the signal step and feature extraction steps. The bandpass filter continues to process the received signal with a value within a certain frequency band. Also, the conversion of the signals to the frequency domain is achieved by fast Fourier transform (FFT). MFDA is a method to determine large and small period deviations in signal types such as EEG (EOG, EMG, ECG). These mentioned methods can be used in signal preprocessing and property extraction steps.
B. CLASSIFICATION
In the classification step, EEG signals with specified features, offline training, or classifications are performed with methods based on neighborhoods and specific signal features of certain classes. The most commonly used classification methods in literature studies are support vector machines, artificial neural networks, K nearest neighbors. Here, the support vector machines (SVMs) utilize boundaries that specify a function in the coordinate plane to separate two or more classes, and this classification involves an offline training step. In addition, artificial neural networks (YSA) with a structure mainly including support vector machines determine input weights according to data sets having output data of a plurality of different inputs, and classify, estimate, etc. It provides a structure that can be used in many places such as. The K nearest neighbors (Knn) place the classes on a particular coordinate plane and make use of the class of the nearest neighbors of the incoming data when doing the classification process.
C. CONTROL OUTPUTS
Different constructions are used in determining the commands for EEG signals and device control operation that are classified after successive steps. These constructs are generally graphical and are expressed in terms of transitions between nodes, just like a finite automata. An example of such a command structure is shown in Fig. In the given structure, a vehicle control operation was performed according to the state of right hand and left hand openness. A given scenario is a random scenario for understanding the structure.
IV. COMPARATIVE RESULTS
As the mentioned earlier when the EEG based control applications in the literature are examined, it is understood that the techniques used in the generalized four steps of EEG and control applications are different. In particular, the step of pre-classification and classification differs in most applications. In addition to these, the classification performances and the type of device controlled are also different. The studies in Table I are compared according to preliminary steps, classification methods and classification performance. In Table II, the user commands to be classified are compared according to the signal frequency ranges obtained as a result of electrode selection and preprocessing of the EEG device.
TABLE I
Some information belongs to literature works.
Refer
ence Preprocessing and Feature Extraction Steps
Classification
Method DeviceType Classification Success [2] IIR Eliptic Band Pass Filter, multi-frequency detrend wave analysis
ANFIS and OVA Robot
Arm %90.11
[3] Band Pass
Filter ANN and SVM Wheelchair %89.88(ANN)%97.1(SVM) [4] - Machine Learning Robot
Arm
-[8] Band Pass Filter- Notch Filter
Machine Learning Mobile Robot -[9] FFT and Gaussian Filter Total Weight of Frequencies Robot -[10] - LDA Virtual Robot %63 [11] CAR(Center Avarage Reference), FFT, Low Pass Filter
Online Classification Robot Arm
%80
[12] FFT, Low
Pass Filter Machine Learning Virtual 3DCube -[13] FFT Threshold Base Classification on SSVEP (Steady State Visually Evoked Potential) signal Robot Arm %100 [14]* Low Pass Filter, FFT, DWT (Discerete Wave Transform)
ANN and SVM Mobile
Robot Training:%100, %90, %80, %90 Test : %95, %90, %80, %85 [16]*
* FIR Band Pass Filter - Robot Arm %95, %73 [18]*
** Band Pass Filter - Robot %52, %95, %55 [19] ICA (Independent Component Analysis) LDA Vehicle %76.4 [20] wCCA (Worst Case Circuit Analysis), CSP(Commo n Space Pattern) KNN Vehicle %84.7
[21] ICA, CSP BPNN (back Vehicle %89.92
propagation neural network)
TABLE III
Some information belongs to literature works.
Reference Brain Activity Used Electrodes Signal Frequency Range Obtained by Preprocessing [3] Right and left hand
activity FC3, FCZ, FC4, C3,C1, CZ, C2, and C4 8 – 30 Hz [5] Thought based
control F3, F4, C3, Cz, C4,P3, Pz, P4 8 - 30 Hz [8] Thought based
control FC3, FCZ, FC4, C3,C1, CZ, C2, and C4 1 - 100 Hz [16] Eye Activity. A1, AF7, AF8, A2 2-5 Hz (AF7,AF8)
2-10 Hz (A1-A2)
When comparing the literature studies, the differences in performance of control applications with different types of classification techniques are linked to the electrode activity used for classification, method of classification, and electrode selection for EEG signals for classification of brain activity . The preprocessing algorithms used in EEG based control applications should be selected in accordance with the focused brain activity. In addition, the selection of the classification method is very important because the classification performances of different classification algorithms are variable for different activities. The EEG-based control is a logical viewpoint for increasing the classification performance of the system in which the different classification methods are compared against the applied problem. In addition, the signal acquisition for the performance of the developed EEG based device control system should be performed with a minimum number of electrodes, which most reveal the focused brain activity. In addition, since the raw signals from these electrodes are quite noisy, operations must be performed at the frequency range that carries the information about the focused brain activity. In applications involving multiple electrode selection, not all signals may be pre-processed at the same frequency range. That is, in an EEG based control application that receives signals from electrodes A1 and F3, the frequency range of the bandpass filter applied to the signals from channel A1 does not have to be equal to the frequency range of the bandpass filter applied to the signals from channel F3. Band filters to be applied in the same frequency range on different channels may affect the performance of the developed system negatively because it produces processed signals with unnecessary range size.
V. CONCLUSION
In this study, performance analysis of the studies focusing on EEG based device control applications in the literature was carried out by considering the different aspects of the studies mentioned and the points that should be considered during the design of such applications are mentioned. EEG based device control applications include the acquisition, preprocessing, feature extraction and classification steps of the EEG signal. In this study, a comparison of preliminary, feature extraction, classification methods and classification performance of device control studies with EEG signals is presented. In addition, electrode selection for EEG signal acquisition,
focused brain activity, and comparison of the signal-to-noise ratio of the bandpass filter used in the preprocessing step. The results of the study are interpreted and the issues that need to be considered in the design of related future related applications are mentioned.
REFERENCES
[1] Gokbulut A. B., Supervisor: Karakose M., Human In The Loop Simulation Framework for Unmanned Aircraft Control through Brain Computer Interface, Master Thesis, December,2017.
[2] Bhattacharyya S., Basu D., Konar A., Tibarewala D. N. Interval type-2 fuzzy logic based multiclass ANFIS algorithm for real-time EEG based movement control of a robot arm, Robotics and Autonomous Systems, 68, pp. 104-115, 2015.
[3] Alomari M. H., Samaha A., AlKamha K., Automated classification of L/R hand movement EEG signals using advanced feature extraction and machine learning. arXiv preprint arXiv: pp.1312.2877, 2013 [4] Ranky G. N., Adamovich S., Analysis of a commercial EEG device
for the control of a robot arm, In Bioengineering Conference, Proceedings of the 2010 IEEE 36th Annual Northeast (pp. 1-2), 2013
[5] Millan J. R., Renkens, F., Mourino J., Gerstner W. Noninvasive
brain-actuated control of a mobile robot by human EEG. IEEE Transactions on biomedical Engineering, 51(6), 1026-1033, 2004.
[6] Jia W., Huang D., Luo X., Pu H., Chen X., Bai O. Electroencephalography (EEG)-based instinctive brain-control of a quadruped locomotion robot, In Engineering in Medicine and Biology Society (EMBC), 2012 Annual International Conference of the IEEE (pp. 1777-1781). IEEE, 2012.
[7] Bi L., Fan X. A., Liu Y., EEG-based brain-controlled mobile robots: a survey, IEEE Transactions on Human-Machine Systems, 43(2), pp.161-176, 2013.
[8] Chae Y., Jeong J., Jo S., Toward brain-actuated humanoid robots: asynchronous direct control using an EEG-based BCI, IEEE Transactions on Robotics, 28(5), 1131-1144,, 2012.
[9] Güneysu A., Akin H. L. , An SSVEP based BCI to control a
humanoid robot by using portable EEG device. In Engineering in Medicine and Biology Society (EMBC), 2013 35th Annual International Conference of the IEEE (pp. 6905-6908). IEEE,2012
[10] Malki A., Yang C., Wang N., Li Z., Mind guided motion control of
robot manipulator using eeg signals, In Information Science and Technology (ICIST), 2015 5th International Conference on (pp. 553-558), IEEE,2012.
[11] Tanwani A. K., & Billard A., Rewards-driven control of robot arm by decoding EEG signals. In Engineering in Medicine and Biology Society (EMBC), 2014 36th Annual International Conference of the IEEE (pp. 1658-1661). IEEE,2014.
[12] Garcia A. P., Schjølberg I., Gale S, EEG control of an industrial robot manipulator, In Cognitive Infocommunications (CogInfoCom), 2013 IEEE 4th International Conference on (pp. 39-44),2013. [13] Perera C. J., Lalitharatne T. D., Kiguchi, K., EEG-controlled meal
assistance robot with camera-based automatic mouth position tracking and mouth open detection, In Robotics and Automation (ICRA), 2017 IEEE International Conference on (pp. 1760-1765), IEEE, 2017, May. [14] Munkanpalli S. V., Sagat, S. P., & Mali, M. D. (2016, December).
Design and development of EEG controlled mobile robots. In Advances in Electronics, Communication and Computer Technology (ICAECCT), 2016 IEEE International Conference on (pp. 133-138). IEEE.
[15] Katona, J., Ujbanyi, T., Sziladi, G., & Kovari, A. (2016, October). Speed control of Festo Robotino mobile robot using NeuroSky MindWave EEG headset based brain-computer interface. In Cognitive Infocommunications (CogInfoCom), 2016 7th IEEE International Conference on (pp. 000251-000256). IEEE.
[16] Minati, L., Yoshimura, N., & Koike, Y. (2016). Hybrid Control of a Vision-Guided Robot Arm by EOG, EMG, EEG Biosignals and Head Movement Acquired via a Consumer-Grade Wearable Device. IEEE Access, 4, 9528-9541.
[17] Kiguchi, K., & Hayashi, Y. (2013, October). Motion estimation based on EMG and EEG signals to control wearable robots. In Systems, Man,
and Cybernetics (SMC), 2013 IEEE International Conference on (pp. 4213-4218). IEEE.
[18] Yordanov, Y., Tsenov, G., & Mladenov, V. (2017, September). Humanoid robot control with EEG brainwaves. In Intelligent Data Acquisition and Advanced Computing Systems: Technology and Applications (IDAACS), 2017 9th IEEE International Conference on (Vol. 1, pp. 238-242). IEEE.
[19] Teng, T., Bi, L., & Fan, X. A. (2015, June). Using EEG to recognize emergency situations for brain-controlled vehicles. In Intelligent Vehicles Symposium (IV), 2015 IEEE (pp. 1305-1309). IEEE. [20] Lu, Y., Hu, Y., Liu, R., Wang, H., Asama, H., & Duan, F. (2016,
July). The design of simulation vehicle system controlled by multichannel EEG based on imaginary movements. In Control Conference (CCC), 2016 35th Chinese (pp. 4976-4981). IEEE. [21] Zhuang, J., & Yin, G. (2017, July). Motion control of a
four-wheel-independent-drive electric vehicle by motor imagery EEG based BCI system. In Control Conference (CCC), 2017 36th Chinese (pp. 5449-5454). IEEE.
[22] Cernea, D., Olech, P. S., Ebert, A., & Kerren, A. (2012). Controlling in-vehicle systems with a commercial eeg headset: performance and cognitive load. In OASIcs-OpenAccess Series in Informatics (Vol. 27). Schloss Dagstuhl-Leibniz-Zentrum für Informatik.
[23] Teng, T., Bi, L., & Liu, Y. (2017). EEG-Based Detection of Driver Emergency Braking Intention for Brain-Controlled Vehicles. IEEE Transactions on Intelligent Transportation Systems.
[24] Utsugi, K., Obata, A., Sato, H., Katsura, T., Sagara, K., Maki, A., & Koizumi, H. (2007, August). Perceptual interaction of optical BCI applications. In ACM SIGGRAPH 2007 posters (p. 155). ACM. [25] Huh, W. G., & Cho, S. B. (2015, November). Optimal partial filters of
EEG signals for shared control of vehicle. In Soft Computing and Pattern Recognition (SoCPaR), 2015 7th International Conference of (pp. 290-293). IEEE.
[26] Krishna, D. H., Pasha, I. A., & Savithri, T. S. (2016, January). Autonomuos robot control based on EEG and cross-correlation. In Intelligent Systems and Control (ISCO), 2016 10th International Conference on (pp. 1-4). IEEE.
[27] Banik, B. C., Ghosh, M., Das, A., Banerjee, D., Paul, S., & Neogi, B. (2017, March). Design of mind-controlled vehicle (MCV) & study of EEG signal for three mental states. In Devices for Integrated Circuit (DevIC), 2017 (pp. 808-812). IEEE.
[28] Senoussi, M., Verdiere, K. J., Bovo, A., Chanel, C. P. C., Dehais, F., & Roy, R. N. (2017, October). Pre-stimulus antero-posterior EEG connectivity predicts performance in a UAV monitoring task. In Systems, Man, and Cybernetics (SMC), 2017 IEEE International Conference on (pp. 1167-1172). IEEE
[29] Singandhupe, A., La, H. M., Feil-Seifer, D., Huang, P., Guo, L., & Li, M. (2017). Securing a UAV Using Individual Characteristics From an EEG Signal. arXiv preprint arXiv:1704.04574.
[30] A.B. Gokbulut, M. Karakose, A Fuzzy Based Brain-Computer Interaction Approach Using Feature Extraction for Personal Identification and Authentication, The 5th International Fuzzy Systems, Symposium (FUZZYSS'17), ABSTRACT, pp. 105, Ankara, Turkey, 14-15 October 2017.
[31] I. Aydin, M. Karaköse, Type-2 Fuzzy Based Quadrotor Control Approach, The 9th IEEE Asian Control Conference (IEEE ASCC 2013), İstanbul, Turkey, 2013.
