The EEG-based Emotion Classification in Tactile, Olfactory, Acoustic

and Visual Modalities

G. Portnova

1,2

, D. Stebakova

and G. Ivanitsky

Institute of Higher Nervous Activity and Neurophysiology of RAS, 5A Butlerova St., Moscow 117485, Russia

Pushkin State Russian Language Institute, Russia

Keywords: EEG, Classification, Acoustic, Visual, Tactile and Olfactory Modalities, Pleasant and Unpleasant Stimuli.

Abstract: We perceive pleasant and unpleasant stimuli using different modality systems, such as visual and acoustic

tactile and olfactory modalities. In our study we investigated the specificity of emotional perception in four

modalities using EEG. 20 healthy participants were instructed to assess the stimuli using emotional scales.

We used power spectrum density, alpha-peak frequency, wavelet analysis and method of "emotional spaces"

for EEG data and DNN classifier for modality specific and non-specific classification of pleasant and

unpleasant stimuli. We found, that difference of EEG power spectrum density and alpha-peak frequency

between states of pleasant and unpleasant stimulation varied from one modality to another. Meanwhile, the

above-stated differences were more similar between tactile and olfactory modalities and acoustic and visual

modalities. the method of "emotional spaces" and DNN classification showed general, modality nonspecific

features of pleasantness evaluation.

1 INTRODUCTION

The perception of emotionally charging stimuli is

possible in variable sensory-specific systems: visual,

auditory, olfactory and tactile, each of them should

be accompanied by the different brain activity (Wu

et al., 2018). Nevertheless, researchers reported, that

the assessment of the "pleasantness" and

"unpleasant" of stimuli should include sensory-non-

specific components (Grabenhorst et al., 2007). The

aim of our study was to detect the modalities’

specific and non-specific features of emotional

perception that could be used for the forehead

classification.

Some researchers previously reported about

similar physiological mechanisms for assessing

emotions in different modalities (Delplanque et al.,

2008). One of these mechanisms could be related

with the activation of the limbic system. The

activation of the limbic system was shown to be

accompanied with theta-rhythm activity (Lévesque

et al., 2017), responsible for emotional perception in

different sensation systems (Diao et al., 2017). Some

authors reported, that the emotional perception of

pleasant and unpleasant stimuli in visual modality

has EEG specific delta- theta- rhythm PSD patterns

(Iosilevich et al., 2012), which was higher for

unpleasant stimulation. Moreover, higher alpha-

rhythm frequency during visual emotional

perception was related with the predisposition to the

prevalence of positive emotions (Tumyalis et al.,

2010). The emotional perception in tactile modality

was also accompanied changes of the theta- and

alpha-rhythm PSD (Monosova, 1994). The pleasant

pleasurable feeling, induced by light pressure that

excites C-tactile fibers, as was shown previously

related with processing of the sensation in limbic

cortical areas (McGlone et al., 2014, McGlown et

al., 2012).

The general mechanisms of emotional perception

originate from asymmetry of pleasant and

unpleasant emotions (Coan and Allen, 2004). The

differential roles of left and right cortex for

processing of pleasant and unpleasant emotional

information was repeatedly reported (Fernandez-

Carriba et al., 2002). Resting EEG measures figure

prominently in this literature. These studies have

established differential roles of left and right

prefrontal cortex (PFC) for processing pleasant and

unpleasant emotional information, respectively. For

example, Loken and co-authors reported that

pleasant tactile stimulation activate left anterior

Portnova, G., Stebakova, D. and Ivanitsky, G.

The EEG-based Emotion Classiﬁcation in Tactile, Olfactory, Acoustic and Visual Modalities.

DOI: 10.5220/0006892100930099

In Proceedings of the 2nd International Conference on Computer-Human Interaction Research and Applications (CHIRA 2018), pages 93-99

ISBN: 978-989-758-328-5

insula, related with the processing of pleasant

emotions (Loken et al., 2009).

Thus, in our study we attempted to investigate

both sensory specific and non-specific mechanisms

of emotional reception and processing using

innovation method of visualization of EEG patterns

(Roik et al., 2014) and Deep neural network

classifier, which was previously reported as effective

tool to detect the emotional states using EEG data

(Stuhlsatz et al., 2011).

2 METHODS

2.1 Subjects

20 healthy right-handed subjects participated in our

study (9 male, 11 female, 30.2±2.7 years old).

Exclusion criteria were: menstrual cycle phase, use of

oral contraceptives, previous neurological or

psychiatric history, pregnancy, treatment with anti-

depressants and anxiolytics and high levels of anxiety

or hostility during the examination (Spielberger et al.,

1970; Buss and Durkee, 1957). Peers have signed the

informed consent for research document indicating

willingness to participate in the study.

2.2 Stimuli

The experiment consisted of 4 series corresponded

to 4 modalities. The quantity of stimuli varied

depending on modality: 16 pictures from IAPS

(Lang, 2008) (6 pleasant, 6 unpleasant, 4 neutral), 12

sound (4 pleasant, 4 unpleasant, 4 neutral), 10 tactile

stimuli (4 pleasant, 4 unpleasant, 2 neutral), 14 odors

(5 pleasant, 5 unpleasant, 4 neutral). Participants

assessed the pleasantness and arousal of stimuli both

during EEG recording (by choosing bottom for most

pleasant (9), neutral (5) and most unpleasant (1), the

gradient was marked on keyboard (1-9)) and after

the experiment using visual scale. Two stimuli (most

pleasant and unpleasant) were selected for farther

classification. All the stimuli’ presentation was

randomized separately for each modality and repeated

4 times for tactile and olfactory modalities (these

stimuli were presented for 24 seconds) and 40 times

for auditory and visual modalities (presented for 8

seconds). The stimuli were presented using Presenta-

tion Software (Neurobehavioral Systems, USA).

2.3 EEG Registration

During the EEG recording the subjects sat in a

comfortable position in an armchair in an

acoustically and electrically shielded chamber. The

participants were instructed to remain calm and to

hear to the presented sounds (via earphones), watch

the visual stimuli (presented in the monitor), smell

the odors, and percept tactile stimuli avoiding falling

asleep. The auditory olfactory and tactile stimuli

were presented while the subject’s eyes were closed,

to avoid visual interference. EEG was recorded

using a recording device Neurotravel-24D (ATES

Medica, Italy) with 32-channel Electro-Cap (USA).

The amplifier bandpass filter was nominally set to

1.6-30 Hz. The electrooculogram (EOG) was

measured with AgCl cup electrodes placed 1 cm

above and below the left eye, and the horizontal

EOG was measured with electrodes placed 1 cm

lateral from the outer canthi of both eyes. The

recording was separated on two datasets with 30-40

minute interruption.

2.4 Data Processing

EEG intervals corresponding to a specific stimulus

were concatenated. These epochs lasting about 300-

400 seconds were analyzed further. Eyes movement

artifacts were cleaned out using EOG data by

EEGLab. Small intervals affected by muscle activity

were excluded (cut) manually using visual

inspection. All the following processing was

performed using EEGLab (Delorme and Makeig,

2004) plugin for MatLab (Mathwork Inc.). The

“emotional spaces” calculations were implemented

on C# programming language by the lab’s engineer.

2.5 Power Spectral Density

Fast Fourier Transform (FFT) was used to analyze

PSD. The EEG spectrum was estimated for each

310±6.8 seconds long interval. The resulting spectra

were integrated over intervals of unit width in the

range of interest (2-2.5Hz, 2.5-3 Hz … 19.5-20 Hz).

We analyzed asymmetry of differences between

pleasant and unpleasant stimuli over symmetric

channels (F7-F8, F3-F4, FC5-FC6, T3-T4, C3-C4,

CP5-CP6, T5-T6, P3-P4, O1- O2), the results were

presented on figure 2.

2.6 Variability of Rhythm

(Wavelet SD)

We applied mathematical method the Morlet

wavelet (or Gabor wavelet). This is a complex

exponential modulated by a Gaussian function

which depends on a tunable parameter is related to

the time and frequency resolutions (Tallon-Baudry

CHIRA 2018 - 2nd International Conference on Computer-Human Interaction Research and Applications

et al, 1996). We calculated the standard deviation for

the intervals of unit width in the range of interest (2-

4Hz, 4-6 Hz, … 18 -20 Hz).

2.7 Peak Alpha Frequency (PAF)

PAF was taken as the frequency from range 8-13 Hz

with maximal PSD.

2.8 Emotional Spaces

We used “cognitive space” construction method

(Roik and Ivanitskii, 2013) to visualize how

close/distant these emotional sound and background

fragments are according to EEG data. As the stimuli

in this study are emotional, the constructed space

will be called “emotional” space. The method

consists of the following steps (figure 1):

1) EEG of each emotional sound and

background fragments was divided into small

non-overlapping epochs of 8 seconds

(approx. 30-40 pieces).

2) FFT (absolute value) was calculated for the

epochs in 2-20 Hz band for electrodes (F3,

F4, F7, F8, FC5, FC6, T3, T4, T5, T6, CP5,

CP6, P3, P4, C3, C4, O1, O2 international

10–20 system)

3) The distance between each pair of emotional

stimuli was calculated: for each frequency

bin two samples of FFT values (of the epochs

of these fragments) were compared using

Mann-Whitney U-test (p < 0.05). The

distance was equal to the percentage of

differing frequency bins.

4) Emotional stimuli were placed onto a plane

using multidimensional scaling method,

namely Sammon projection (Sammon, 1969).

Each type of modality is depicted using the shape

the pleasant and unpleasant stimuli were depicted

using color (see Figure 3 A). So, the distances

between the stimuli types on the plane were as

similar as possible to the distances calculated by

FFT values. This similarity was always good enough

to claim the projection is legit.

5) The resulting pictures (obtained for each

subject) have arbitrary rotation because of

Sammon projection algorithm and different

sizes because of high individuality of EEG.

Before the averaging over group these

pictures should be standardized. We used

scaling to equalize the size (the sum of

squared distances to the figures from the

“center of mass”) and rotation/reflection so

that pleasant visual stimulus (white rhomb)

was on the top of the picture and the

unpleasant and pleasant auditory stimuli

(circles) were on the left and right sides

correspondingly laying on a horizontal line.

6) After standardization individual pictures are

averaged over groups. So, these pictures

show relational distances between emotional

sounds based on how much the

corresponding EEG data differ in terms of

rhythms magnitudes.

Figure 1: Steps of “Cognitive spaces” method.

2.9 Statistical Analysis

A one way ANOVA with Bonferroni correction for

multiple comparisons, p < 0.05, were used to

determine lateralization effects on EEG metrics. We

analyzed differences of EEG distances using

Student's t-test to compare indices for each stimulus

(p < 0.05). The Pearson’s correlation coefficient

between EEG indices and emotional assessments

was calculated. Significant R values were used for

further analysis (p < 0.05).

2.10 Classifier

The Deep neural network (DNN) and Extreme

Learning Machines (ELM) was used for the

classification of pleasant and unpleasant stimuli

recognition using the EEG signals (Han, 2014;

Tripathi, 2017). The testing sample was taken from

the dataset 1 and then passed on to the trained

The EEG-based Emotion Classiﬁcation in Tactile, Olfactory, Acoustic and Visual Modalities

network. We had three data arrays, which contained

from 32 different channels: power spectral density,

alpha-peak frequency, and wavelet data. Wе

prepared two datasets

2.10.1 Dataset I

EEG data was taken from a first part of study, when

subjects assessed the pleasantness of different

stimuli. After the first type of EEG study subjects

assessed the stimuli using psychological scales. The

most “pleasant” and “unpleasant” stimuli was

selected using self-reported assessment and

psychometric scales and divided in two groups:

training and testing.

The classifier was trained on two types of EEG

data: 1) using most pleasant and unpleasant stimuli

separately for different modalities (8 groups,

sensory-specific) 2) using most pleasant and

unpleasant stimuli averaged over all modalities (2

groups, sensory-non-specific).

2.10.2 Dataset II

EEG datasets were taken from the second part of

study, when subjects were instructed as previously.

The tested EEG data contained pleasant and

unpleasant stimuli with the similar emotional

characteristics. The percentage of correct

classification was measured for each subject

separately.

2.11 Emotional Assessment of Stimuli

After the first part of the EEG registration subjects

were instructed to assess stimuli using specially

prepared questionnaire. The questionnaire included:

specification of presented stimuli and several scaled,

measuring emotional features (“Pleasantness”,

“Fear”, “Arousal”, “Disgust” and etc.) Participants

were instructed to indicate how the stimuli describe

their affective state on a scale from 0 (“not at all”) to

5 (“extremely”).

3 RESULTS

3.1 Power Spectral Density

The rhythmic spectral activity of more ancient

modalities (tactile and olfactory) was differed from

more modern modalities (auditory and visual): the

slow-wave rhythm PSD was lower and beta-rhythm

BSD was higher for ancient sensory systems

(p<0.05).

The differences of PSD between pleasant and

unpleasant stimuli showed significant asymmetry

(Figure 2). In the right hemisphere we found

significant differences of PSD between pleasant

stimuli for each modality type. In the left

hemisphere only visual and olfactory pleasant and

unpleasant stimuli’ PSD had significant differences.

The visual pleasant stimuli (compared to unpleasant)

had lower delta-rhythm PSD in the left hemisphere

and higher alpha-and beta-rhythm bilateral. The

auditory pleasant stimuli had lower delta and theta-

rhythm PSD and higher beta-rhythm in the right

hemisphere. The tactile pleasant stimuli had higher

alpha- and beta-rhythm in the right hemisphere. The

olfactory pleasant stimuli had higher alpha- and

beta-rhythm in the right hemisphere and lower beta-

rhythm in the left hemisphere.

Figure 2: The differences of the PSD in the right (A) and

left (B) hemisphere between pleasant and unpleasant

stimuli.

3.2 Alpha-peak Frequency

The alpha-peak frequency was significantly higher

for unpleasant stimuli compared to pleasant in the

right central and temporal areas (C4, T4, F8, Cz,

Pz). These differences were found for tactile,

auditory and visual stimuli.

CHIRA 2018 - 2nd International Conference on Computer-Human Interaction Research and Applications

3.3 Wavelet Standard Deviation

Significant differences were found only for visual

stimuli. Standard deviation was significantly higher

for pleasant visual stimuli: for theta- and delta-

rhythm in the right temporal areas (F8,T4,T6) and

for the alpha- and beta-rhythm in the central and

parietal areas (F8, F4, C4,Cz, C3 T4, T6, T5 Pz, P3,

O1).

3.4 Classification

The results of sensory-specific and sensory non-

specific classification are presented in Table 1.

Table 1: The percentage of correct classification (averaged

over the group).

DNN for 8 classes

Pleasant Unpleasant

Visual

0.81±0.02 0.84±0.05

Auditory

0.78±0.06 0.85±0.07

Tactile

0.80±0.01 0.71±0.05

Olfactory

0.88±0.04 0.89±0.03

Sensory-non-specific

DNN for 2 classes

Pleasant Unpleasant

Visual

0.65±0.01 0.71±0.04

Auditory

0.64±0.06 0.61±0.02

Tactile

0.72±0.02 0.58±0.01

Olfactory

0.69±0.03 0.76±0.05

Sensory-non-specific

0.64±0.07 0.67±0.04

3.5 Emotional Spaces

EEG differences represent the both sensory-specific

and sensory-non-specific differences between

stimuli (Figure 3). The unpleasant stimuli for each

modality were in the left side of the emotional space,

compared to pleasant stimuli. The more ancient

sensory systems were separated from the more

modern sensory systems. EEG distances between

pleasant and unpleasant stimuli positively correlated

with the distances of emotional assessment by the

scale “Pleasantness” (r>0.48, p<0.05). To calculate

distances of emotional assessment we analyzed

difference between scores of pleasant and unpleasant

stimuli.

Figure 3: A: “emotional spaces”, B: the subjective

assessment of stimuli using scales “Arousal” and

“Pleasantness”.

4 DISCUSSION

In our study we found that modern sensory systems

(visual and auditory) had similar EEG patterns and

differed from more ancient sensory systems

(olfactory and tactile). In spite of a small amount of

data, which analyzed emotional perception in four

different modalities simultaneously, some

researchers reported about similarity of the EEG

rhythmic activity in alpha- and beta-bands between

visual and auditory systems (Jessen and Kotz, 2011).

These data correspond to our results showed similar

beta-rhythm PSD between pleasant and unpleasant

stimuli in visual and auditory modalities.

We hypothesized that modality-independent

mechanisms of emotional processing always

accompany the emotional perception of pleasant and

unpleasant stimuli; the results of our study seem to

confirm this assumption. For example, we found the

good level of classification accuracy trained on

sensory-non-specific EEG distances. This modality-

independent difference between pleasant and

unpleasant stimuli also could be visualized using

“Emotional spaces” method. Other researchers also

reported about modality non-specific emotional

stimuli processing, which occurs when subjects

solve tasks, presented in different sensory systems

(Brosch, 2009). Previous research has demonstrated

that emotions from faces and emotions from voices

are also represented using similar mechanisms, for

example, both types of emotional stimuli have been

shown to be processed in the superior temporal

sulcus (Haxby et al., 2002; von Kriegstein and

Giraud, 2004).

Our results also demonstrated the asymmetry of

EEG changes during of emotional perception. This

data is consistent with the previous studies reported

about the brain asymmetry during processing of

The EEG-based Emotion Classiﬁcation in Tactile, Olfactory, Acoustic and Visual Modalities

pleasant and unpleasant stimuli and hypothesized

that positive emotions correspond to the right

hemisphere, and negative – to the left (Fernandez-

Carriba et al., 2002). For example, the emotion-

modulated asymmetries, related with processing

pleasant and unpleasant emotional information were

found in the frontal cortex (Coan and Allen, 2004).

The clinical EEG studies have shown that depression

is associated with the greater activation of the right

prefrontal cortex (Davidson et al., 2002), other

researchers also reported about the higher activation

of the right amygdala (Abercrombie et al., 1998).

Furthermore, our results showed that most

pronounced differences of the EEG between

pleasant and unpleasant stimuli were found in the

right hemisphere. Previously, a general right

hemispheric advantage for emotion processing was

reported (Martin and Altarriba, 2017; Kesler-West et

al., 2001).

5 CONCLUSIONS

Visual and auditory sensory systems had similar

EEG patterns and differed from olfactory and tactile

sensory systems. The good level of classification

accuracy trained on sensory-non-specific EEG

distances was found. The advantage of the right

hemisphere for emotional processing was found. The

modality-independent difference between pleasant

and unpleasant stimuli is primarily visualized with

the “Emotional spaces” method. Further work is

needed to be done with the increased number of

healthy participants. Moreover, we are going to

include the patients with emotional impairments in

our study. The techniques used for classification

should be extended to support reported findings

ACKNOWLEDGEMENTS

We would like to thank engineer Kashevarova O,

researchers Atonov M, and Portnov V for assistance

in programming of “Cognitive spaces”, calculations

of the EEG parameters and DNN + EML

classification

REFERENCES

Abercrombie H, Schaefer S, Larson C, Oakes T, Lindgren

K, Holden J, et al. , 1998 .Metabolic rate in the right

amygdala predicts negative affect in depressed

patients. In Neuroreport, 9(14):3301–3307.

Brosch T, Grandjean D, Sander D., Scherer K. R. 2009.

Cross-modal emotional attention: emotional voices

modulate early stages of visual processing. J Cogn

Neurosci. 21(9):1670-9

Buss A, H,, Durkee A., 1957. Inventory for assessing

different kinds of hostility.In J Consult Psychol

21:343–9.

Coan J, Allen J., 2004. Frontal EEG asymmetry as a

moderator and mediator of emotion. In Biological

Psychology, 67:7–49

Davidson R, Pizzagalli D, Nitschke J, Putnam K., 2002.

Depression: Perspectives from affective neuroscience.

In Annual Review of Psychology;53(1):545–574.

Delorme, A., Makeig, S. 2004. EEGLAB: an open source

toolbox for analysis of singletrial EEG dynamics

including independent component analysis. J.

Neurosci. Methods, 134, 9–21.

Delplanque S, Grandjean D, Chrea C, Aymard L, Cayeux

2008. I, Le Calve B, Velazco MI, Scherer KR, Sander

D. Emotional processing of odors: evidence for a

nonlinear relation between pleasantness and

familiarity evaluations. Chemical Senses. Apr

9;33(5):469-79.

Diao, L., Qi, S., Xu, M., Fan, L., and Yang, D. 2017.

Electroencephalographic theta oscillatory dynamics

reveal attentional bias to angry faces. In Neuroscience

letters, 656, 31-36.

Fernandez-Carriba, S., A. Loeches, A. Morcillo, and W.D.

Hopkins, 2002. Functional asymmetry of emotions in

primates: new findings in chimpanzees. In Brain

Research Bulletin. 57: 561–564.

Han, K., Yu, D. and Tashev, I., 2014. Speech emotion

recognition using deep neural network and extreme

learning machine. In Fifteenth annual conference of

the international speech communication association.

Haxby, J., Hoffman, E., and Gobbini, I., 2002. Human

Neural Systems for Face Recognition and Social

Communication. In Biological Psychiatry. 51(1):59-

Grabenhorst, F., Rolls, E. T., Margot, C., da Silva, M. A.

and Velazco, M.I., 2007. How pleasant and unpleasant

stimuli combine in different brain regions: odor

mixtures. Journal of Neuroscience, 27(49), pp.13532-

13540.

Iosilevich E. A., Chernysheva E. G., Chernyshev B. V.,

2012. Psychophysiological study of the connection

between the valence of the emotional response and the

EEG spectral power indicators of human. In: Modern

Psychology: Theory and Practice: Materials of the V

International Scientific and Practical Conference,

Moscow, July 3-4, M.: Special book, 2012. S. 21-27.

(in Russian)

Jessen S, Kotz S. A., 2011. The temporal dynamics of

processing emotions from vocal, facial, and bodily

expressions. In Neuroimage. Sep 15;58(2):665-674

Kesler-West, M., Andersen, A., Smith, C., Avison, M.,

Davis, C., Kryscio, R., Blonder, L. 2001. Neural

substrates of facial emotion processing using fMRI.

Cognitive Brain Research, 11(2):213-26

CHIRA 2018 - 2nd International Conference on Computer-Human Interaction Research and Applications

Lévesque, M., Cataldi, M., Chen, L. Y., Hamidi, S. and

Avoli, M., 2017. Carbachol-induced network

oscillations in an in vitro limbic system brain slice.

Neuroscience, 348, pp.153-164.

Löken, L. S., Wessberg, J., McGlone, F., and Olausson, H.

2009. Coding of pleasant touch by unmyelinated

afferents in humans. Nature neuroscience, 12(5), 547-

548.

Martin, J. M. and Altarriba, J., 2017. Effects of valence on

hemispheric specialization for emotion word

processing. Language and speech, 60(4), pp.597-613.

McGlone F., Olausson H., Boyle J. A., Jones-Gotman M.,

Dancer C., Guest S., et al. 2012. Touching and feeling:

differences in pleasant touch processing between

glabrous and hairy skin in humans. Eur. J. Neurosci.

35, 1782–1788

McGlone F, Wessberg J and Olausson H, 2014.

Discriminative and Affective Touch: Sensing and

Feeling. In Neuron, 82, May 21, pp. 738-755

Monosova A. J. 1994. PhD theses. Analysis of the features

of the emotional sphere in normal and affective

disorders by olfactory stimulation. M, (in Russian)

Roik A. O, Ivanitskii G. A., 2013. Neurophysiological

Model of the Cognitive Space //Neuroscience and

Behavioral Physiology. V 43(2):193-199.

Roik A. O., Ivanitskii G. A., Ivanitskii A. M. 2014. The

Human Cognitive Space: Coincidence of Models

Constructed on the Basis of Analysis of Brain

Rhythms and Psychometric Measurements. Neuro-

science and Behavioral Physiology. 44 (6): 692-701.

Sammon J. W. 1969 A nonlinear mapping for data

structure analysis //IEEE Transactions on computers.

V 18(5): 401-409.

Spielberger C. D., Goursch R. L. and Lushene R. E. 1970

Manual for the State-Trait Anxiety Inventory.

Consulting Psychologist Press, Palo Alto, CA

Stokes, D., Matthen, M., and Biggs, S. (Eds.). 2014.

Perception and its modalities. Oxford University

Press, USA.

Stuhlsatz, A., Meyer, C., Eyben, F., Zielke, T., Meier, G.,

and Schuller, B. 2011. Deep neural networks for

acoustic emotion recognition: raising the benchmarks.

In Acoustics, speech and signal processing (ICASSP).

IEEE international conference on (pp. 5688-5691).

Tallon-Baudry, C., Bertrand, O., Delpuech, C. and Pernier,

J., 1996. Stimulus specificity of phase-locked and non-

phase-locked 40 Hz visual responses in human.

Journal of Neuroscience, 16(13), pp.4240-4249.

Tripathi, S., Acharya, S., Sharma, R.D., Mittal, S. and

Bhattacharya, S. 2017. Using Deep and Convolutional

Neural Networks for Accurate Emotion Classification

on DEAP Dataset. In AAAI, pp. 4746-4752.

Tumyalis, A. V., Korenkov V. V., Brak I. V., Makhnev V.

P., Reva N. V., Aftanas L. I. 2010. The Individual

frequency of alpha activity and emotions negative

positive emotions / / Bulletin RAS, V. 30(4).

Von Kriegstein, K. and Giraud, A.-L. 2004. Distinct

functional substrates along the right superior temporal

sulcus for the processing of voices. NeuroImage,

22(2), 948-955

Wu, Y. H., Uluç, I., Schmidt, T. T., Tertel, K., Kirilina, E.,

and Blankenburg, F. 2018. Overlapping frontoparietal

networks for tactile and visual parametric working

memory representations. NeuroImage, 166, 325-334.

APPENDIX

The work was supported by the grant of Russian

Foundation for Basic Research № 16 – 04 -00092 A

and the Russian Academy of Science.

The EEG-based Emotion Classiﬁcation in Tactile, Olfactory, Acoustic and Visual Modalities