Page 56 - IJEEE-2023-Vol19-ISSUE-1
P. 56

52 |                                                                                     Al-Qaysi1, Al-Saegh, Hussein & Ahmed

                 IV. RESULTS AND DISCUSSION                        deployment of the proposed framework model in MI-based
                                                                   BCI applications.
      It was mentioned that this study uses CWT to transform
the given MI EEG signals into scalograms (time-frequency                                     V. CONCLUSION
images). Different mother wavelet functions produce
different time-frequency characteristics. To examine those              Deep learning and wavelet transformation are useful
differences, three types of mother wavelets namely Morlet,         techniques for dealing with the high-dimensional and
Bump, and Amor were used in this study.                            nonstationary MI EEG signals. This paper studied the use of
                                                                   deep learning, three different wavelet mother functions, and
     Also, the hybrid feature extraction and classification        six different classifiers for the analysis of MI EEG signals.
model has been achieved by using the VGG-16 as a features          The performance of different combinations of mother
extractor with one classifier. Six classification algorithms       functions and classifiers are compared on two MI EEG
have been examined namely neural network (NN), K-nearest           datasets. Several evaluation metrics show that a model of
neighbors (KNN), Naïve Bayes (NB), logistic regression             VGG-16 feature extractor with a neural network classifier
(LR), SVM, and decision tree (DT).                                 using the Amor mother wavelet function has outperformed
                                                                   the results of state-of-the-art studies by achieving 99% of
     The six classifiers are experimented with each of the         classification accuracy on dataset-II and 100% accuracy on
mother wavelet functions to find the best hybrid model over        dataset I. This result will facilitate the deployment of an
the combined subject dataset of (BCI Competition dataset 2b        accurate model based on our technique to help the
, the training part). The results of this experiment are           community of the BCI users.
presented as confusion matrices in Fig. 4. A confusion matrix
is presented for each classifier and mother wavelet function.                           CONFLICT OF INTEREST
Fig. 5 shows the region of convergence (ROC) for each one
can see that NN based hybrid model outperformed the                     The authors have no conflict of relevant interest to this
performance of other classification methods with the three         article.
mother wavelet functions.
                                                                                              REFERENCES
     It can be noticed that CNN+NN has achieved the best
results in comparison to the other experimented hybrid             [1] A. Al-Saegh, et al., "Deep learning for motor imagery
models. The confusion matrix shows 100% classification               EEG-based classification: A review," Biomedical Signal
accuracy with Amor and Morlet wavelet mother functions.              Processing and Control, vol. 63, p. 102172, 2021.
For this optimal model, the training time, testing time, and
log loss are computed for the three mother functions. As           [2] Z. Al-Qaysi, et al., "A review of disability EEG based
shown in Fig. 6 and overall, the WTNN model with Amor                wheelchair control system: Coherent taxonomy, open
mother functions has delivered the best classification results.      challenges and recommendations," Computer methods and
                                                                     programs in biomedicine, vol. 164, pp. 221-237, 2018.
     The second experiment is depicted for evaluating the
proposed model with another different MI EEG dataset               [3] S. Sreeja and D. Samanta, "Classification of multiclass
which is (dataset II training part, but for individual subject) .    motor imagery EEG signal using sparsity approach,"
This helps overcome the inter-subject classification problem.        Neurocomputing, vol. 368, pp. 133-145, 2019.
To evaluate the optimal model (the WTNN) over different
brain signal complexities to overcome the problem of inter-        [4] S. U. Amin, et al., "Deep Learning for EEG motor
subjects, the model was tested in experiment-2 with another          imagery classification based on multi-layer CNNs feature
dataset (dataset II evaluation part, individual subjects) that       fusion," Future Generation computer systems, vol. 101,
consists of nine subjects. The result showed that the                pp. 542-554, 2019.
developed model attained 99% of mean accuracy over the
nine subjects as presented in TABLE II. Additionally, to           [5] S. Aggarwal and N. Chugh, "Signal processing
evaluate the performance of the WTNN in the ability to               techniques for motor imagery brain computer interface: A
overcome intra- subjects' brain signal challenges of sessions        review," Array, vol. 1, p. 100003, 2019.
(in this study we consider the problem of with feedback and
without feedback recording protocols of two sessions). The         [6] X. Tang, et al., "Motor imagery EEG recognition with
result showed that the WTNN model attained 99% of mean               KNN-based smooth auto-encoder," Artificial intelligence
accuracy over the nine subjects as presented in TABLE III.           in medicine, vol. 101, p. 101747, 2019.
Comparing the result of this study with academic literature
over dataset-I and dataset-II, it is clear that our WTNN model     [7] A. Al-Saegh, et al., "CutCat: An augmentation method
outperformed the accuracy of the literature as presented in          for EEG classification," Neural Networks, vol. 141, pp.
TABLE IV and TABLE IV. This comparative result                       433-443, 2021.
appraises the efficiency of the proposed WTNN model due
to the capability of VGG-16 and Amor wavelet in extracting         [8] C. H. Nguyen and P. Artemiadis, "EEG feature
MI signal features. And also, the efficiency of the hybrid           descriptors and discriminant analysis under Riemannian
model in decoding the right and left commands. This model            Manifold perspective," Neurocomputing, vol. 275, pp.
will contribute to the BCI community by facilitating the             1871-1883, 2018.

                                                                   [9] W. Qiao and X. Bi, "Ternary-task convolutional
                                                                     bidirectional neural turing machine for assessment of
                                                                     EEG-based cognitive workload," Biomedical Signal
                                                                     Processing and Control, vol. 57, p. 101745, 2020.
   51   52   53   54   55   56   57   58   59   60   61