/AT82.01-brain-project-group-4

Primary LanguageJupyter Notebook

AT82.01-brain-project-group-4

Overall guide of files and folders

  • Main presentation file is CP Group 4 Project (MAIN).pdf
  • Each folder with the members name contains the
    • code : notebook of the experiments
    • papers read : Reference papers for the experiments

Progress reports

Week 1 Progress

Completed

  • TOPIC SELECTED: Val, Dom, Arousal from DEAP dataset
  • Analyze and Explore DEAP dataset
  • Reading Papers
    • The Impact of Different Sounds on Stress Level in the Context of EEG, Cardiac Measures and Subjective Stress Level: A Pilot Study (Sunny)
    • Finding Emotion Dataset DEAP dataset (Jirasak , Arnajak)
    • DEAP : A database for Emotion Analysis using Physiological Signals (Sunny, Jirasak, Arnajak)
    • EEG Signals to Measure Mental Stress (Arnajak)
    • Emotion Recognition Based on EEG using DEAP Dataset (Jirasak)

Week 2 Progress

In-Progress

  • Understanding EEGNet and MNE tool(Arnajak)

  • Designing Possible model for DEAP (Jirasak)

  • Feature/ Information (preprocessing) Extraction from DEAP dataset (Sunny, Jirasak)

  • Reading Papers -

    • EEG-Based Emotion Classification Using a Deep Neural Network and Sparse Autoencoder (Jirasak)
    • EEG-based Brain-Computer Interfaces (BCIs): A Survey of Recent Studies on Signal Sensing Technologies and Computational Intelligence Approaches and Their Applications
    • EEG-based emotion recognition in music listening: A comparison of schemes for multiclass support vector machine (Amanda)

Week 3 Progress

Completed

  • Propose and demonstrate on using Welch's method to extract feature from data (Sunny)
  • Predict the Val, Dom, Arousal using 4 bands of area under the curve of Welch's method transformed data from each channel, using simple fully-connected deep neural network (Jirasak, Sunny)
  • Try using EEGNet on the data (Arnajak)

In-Progress

  • Consider Gabor Transform for turning raw data to 3D spectrography, in order to retain temporal feature (Jirasak)
  • Look into Data cleaning/preprocessing step (ICA, etc.) to reduce unnecessary noise (Arnajak)
  • Trying out EEGNet and its derivatives on the DEAP dataset directly (Sunny)
  • Reading the paper on EEGNET (Sunny)

Week 4 Progress

Completed

  • Read 3 papers and Upload paper's PDF files and their summary in github/google drive (Jirasak)
  • Read new paper and uploaded the summary on google drive. (Amanda)

Week 5 Progress

Completed

  • Read 2 paper , Upload paper summary in google drive and Do LSTM model but stuck in Overfitted problem. (Arnajak)
  • Creating scalogram using continuous wavelet transform to feed into CNN network.

Week 6 progress

  • Creating STFT + CNN Network for DEAP Valence and Arousal (Sunny)
  • Tried PSD with 4 bands on 32 channels. Created 128 combined features and ran it on ML, (NVB ,SVC-rbf, KNN ,XGBoost) (Sunny)
  • face problem overfitting or maybe noise in both of model ( psd->LSTM ) and (STFT -> conv2d) (Arnajak)
  • Successfully replicate a model using continuous wavelet transform from this github, CWT preprocess code and scale selection code also translated from matlab to python (Jirasak)
  • Creating STFT + LSTM model for Valence, Arousal, Dominance analysis (Amanda)

st122162 - Jirasak Buranathawornsom
st122458 - Arnajak Tungchoksongchai
st122336 - Sunny Kumar Tuladhar
st122245 - Amanda raj Shrestha