Deep Learning based Brain Activity recognition using EEG.pptx

Visual stimuli trigger neural spikes sent to the brain.
EEG signals analyze stimuli for object detection and classification.
Vision imagery (VI) mental tasks control Brain Computer Interface design.
Enables interaction for locked-in individuals.
Deep learning surpasses traditional methods of manual feature extraction.
CNNs applied to classify stimuli-evoked EEG signals

1. Deep Learning Framework for Recognizing Brain Activity from Character Identification using EEG

2. Outline :- Introduction Work Done in Previous Semester Work done in Current Semester Conclusion References

3. Introduction Visual stimuli trigger neural spikes sent to the brain. EEG signals analyze stimuli for object detection and classification. Vision imagery (VI) mental tasks control Brain Computer Interface design. Enables interaction for locked-in individuals. Deep learning surpasses traditional methods of manual feature extraction. CNNs applied to classify stimuli-evoked EEG signals

4. Brain Computer interface (BCI) BCIs utilize various methods to detect and interpret brain signals. I nvasive methods ( intracortical , ECoG , etc ) Non-invasive methods(MRI, EEG, etc ). Enables communication between the brain and an external device. Allows individuals to control devices or interact with software using their brain activity alone. Some of the applications of BCI are: Assistive technology Rehabilitation Gaming and entertainment Cognitive and mental health applications.

5. Brain Computer interface (BCI) cont … Motivation Empowering individuals with motor disabilities . Bridging the communication gap. The analysis of EEG signals with a larger number of classes along with very large instances of EEG datasets.

6. Brain Computer interface (BCI) cont … Objectives : Design and development of Robust character/numeric digit based EEG classification methodology.

7. Dataset Datasets Group Task during EEG recording Number of Subjects Electrodes Luis Carlos Sarmiento [3] Imagined Speech (vowels) Imagine , Relax 50 14 MindBigData [ EPOC] [10] Seeing and Thinking (numbers) Seeing, Thinking, Relax 1 14 Table 1: Available Datasets

8. Work Done in Previous Semester Fig. 1: Attention mechanism based Architecture

9. Results Method Accuracy(%) FPR(%) FNR(%) F1 Score(%) Proposed 71.333 5.304 11.227 71.330 Spatial and temporal attentions are incorporated into CNN to learn global dependencies. Model yields an average performance in promoting the EEG decoding. Improvement in the accuracy and reducing the number of calculated parameters will be the future objective. Table 2: Result of attention based model

10. Work in C urrent Semester Signal Processing Feature Extraction Classification Fig. 2 : Architecture of Proposed Model

11. Work in the current semester (cont..) Fig 3: Classification Model [4]

12. Layers Parameters Activation Con1D Filter 256/Kernel 7 ReLU BN Maxpooling 2 Con1D Filter 128/Kernel 7 ReLU BN Maxpooling 2 Con1D Filter 64/Kernel 7 ReLU BN Maxpooling 2 Con1D Filter 32/Kernel 7 ReLU BN Maxpooling 2 Flatten Dense 128 ReLU Dense 64 ReLU Dense 10 Softmax Table 3 : Model summary for 1D CNN Model

13. Accuracy Precision Recall F1-Score 98.27 98.05 97.56 97.8 Table 4 : Paper Result [4] in percentage Table 5: Our implementation result in percentage Accuracy Precision Recall F1-Score 97.73 97.44 96.38 96.9

14. Proposed Architecture Fig. 4: Proposed Architecture

15. Table 6: Proposed model result in percentage Optimizer Loss Accuracy Precision Recall F1-Score Adam Train 5.45 98.20 98.39 98.06 98.20 Test 728.86 9.63 9.48 8.77 9.56 SGD Train 203.42 27.08 64.99 1.93 26.84 Test 255.32 10.00 10.00 00.14 8.80 Nadam Train 5.16 98.28 98.45 98.14 98.28 Test 713.12 10.81 10.76 9.95 10.76 AdaGrad Train 140.52 50.90 71.35 29.51 50.84 Test 319.68 10.63 11.13 3.18 10.56 Learning Rate = 0.001 Batch Size = 32

16. Fig. 5 : Metrics visualization with Adam Optimizer Fig. 6: Metrics visualization with SGD Optimizer

17. Fig. 7: Metrics visualization with Nadam Optimizer Fig. 8 : Metrics visualization with AdaGrad Optimizer

18. Conclusion & Future work Training accuracy of the proposed model with Adam and Nadam optimizers are good. Testing/validation accuracy is not good because model is over-fitting. Improvement in the testing/validation accuracy by removing the over- fitting as a future objective. Reducing the number of calculated parameters will be the future objective .

19. Timeline Objective 1: Design and development of Robust character based EEG classification methodology . (Expected date of completion : Dec 2023) Objective 2: Design and development of system for BCI c haracter identification using EEG. (Expected date of completion : Jun 2024) Objective 3: Multi modality based non-invasive BCI system for c haracter identification using EEG . ( Expected date of completion : Dec 2024) Objective 4: Design and development of wearable device for c haracter identification using EEG . ( Expected date of completion : Jun 2025)

20. References [ 1 ] Rami Alazrai , Motaz Abuhijleh , Mostafa Z. Ali, Mohammad I. Daoud , A deep llearning approach for decoding visually imagined digits and letters using time–frequency–spatial representation of EEG signals, Expert Systems with Applications, Volume 203, 2022, 117417, ISSN 0957-4174, 10.1016/j.eswa.2022.117417 . [2] Zhao Y, Chen Y, Cheng K, Huang W. Artificial intelligence based multimodal language decoding from brain activity: A review. Brain Res Bull. 2023 Sep;201:110713. doi : 10.1016/j.brainresbull.2023.110713. Epub 2023 Jul 23. PMID: 37487829. [3] K . Wahengbam , K. L. Devi and A. D. Singh, "Fortifying Brain Signals for Robust Interpretation," in IEEE Transactions on Network Science and Engineering, vol. 10, no. 2, pp. 742-753, 1 March-April 2023, doi : 10.1109/TNSE.2022.3222362 . [4] Tiwari , S., Goel , S. & Bhardwaj , A. EEG Signals to Digit Classification Using Deep Learning-Based One-Dimensional Convolutional Neural Network.  Arab J Sci Eng   48 , 9675–9691 ( 2023), doi : 10.1007/s13369-022-07313-3. [5] Ullah S, Halim Z. Imagined character recognition through EEG signals using deep convolutional neural network. Med Biol Eng Comput . 2021 May;59(5): 1167-1183, doi : 10.1007/s11517-021-02368-0. Epub 2021 May 4. PMID: 33945075 . [6] Mishra , Alankrit , Nikhil Raj and Garima Bajwa . “EEG-based Image Feature Extraction for Visual Classification using Deep Learning.”  2022 International Conference on Intelligent Data Science Technologies and Applications (IDSTA)  (2022): 181-188 . [7] Zhao Y, Chen Y, Cheng K, Huang W. Artificial intelligence based multimodal language decoding from brain activity: A review. Brain Res Bull. 2023 Sep;201:110713. doi : 10.1016/j.brainresbull.2023.110713. Epub 2023 Jul 23. PMID: 37487829.

21. References [8]Sarmiento LC, Villamizar S, López O, Collazos AC, Sarmiento J, Rodríguez JB. Recognition of EEG Signals from Imagined Vowels Using Deep Learning Methods. Sensors (Basel). 2021 Sep 29;21(19):6503. doi : 10.3390/s21196503. PMID: 34640824; PMCID: PMC8512781 . [9]Nitta Tsuneo, Horikawa Junsei , Iribe Yurie , Taguchi Ryo, Katsurada Kouichi , Shinohara Shuji , Kawai Goh,Linguistic representation of vowels in speech imagery EEG, Frontiers in Human Neuroscience,VOL . 17, 2023, 10.3389/fnhum.2023.1163578 ISSN: 1662-5161. [10] Panachakel , J.T., Ramakrishnan , A.G. and Ananthapadmanabha , T.V., 2020. A novel deep learning architecture for decoding imagined speech from EEG. arXiv preprint arXiv:2003.09374.

22. THANK YOU