edgarmoyete / deep-learning-bci-iv-2a

Clasificacion de imagenes motoras en señales EEG con CNN, LSTM y otros clasificadores

Deep-Learning-BCI-IV-2a

Clasificación de imaginación motora en señales de EEG con Deep Learning y Machine Learning utilizando BCI Competition IV dataset 2a

Modelos para RAW, DWT-Coef, AlfaC3C4 y RWE-DWT:

• CNN-2D
• CNN-1D
• LSTM
• GRU
• CNN-1D + LSTM
• CNN-1D + GRU
• ConvLSTM2D
• MLP
• SVM
• Random Forest

Modelos para Espectros-STFT y Escalogramas-CWT

• CNN-2D
• CNN-2D + LSTM
• CNN-2D + GRU
• ConvLSTM2D
• CNN-3D

Data Set:

• BCI Competition IV-2a

RAW

• Datos crudos, 22 canales con 1000 valores (4seg a 250Hz), filtrado con pasa banda (8Hz-30Hz)
• Input Shape: 288x22000

RWE-DWT

• RWE con db4 y 2 niveles de descomposicion, 22 canales con 3 valores
• Input Shape: 288x66

AlfaC3C4

• Alfa (8Hz-13.5Hz), 2 canales (C3 y C4) con 255 valores
• Input Shape: 288x510

DWT-Coef

• Coeficientes de aproximacion y detalle concatenados de DWT-db4 de 2 niveles de descomposicion, 22 canales con 1013 valores
• Input Shape: 288x22286

Espectros-STFT

• Espectrogramas con la transformada de Fourier de tiempo corto, ventana de tukey de 0.25

• STFT-CNN.ipynb

• STFT-CNN-CV.ipynb

• STFT-CNN-GS.ipynb

• STFT-CNN-Predecir.ipynb

• STFT-CNN-LSTM.ipynb

• STFT-CNN-LSTM-CV.ipynb

• STFT-CNN-LSTM-GS.ipynb

• STFT-CNN-LSTM-Predecir.ipynb

• STFT-CNN-Explorar.ipynb

• Fine-Tuning.ipynb

• STFT3D-CNN2D-2C-V.ipynb

• STFT3D-CNN2D-2C-V-Resize.ipynb

• STFT3D-CNN2D-H.ipynb

• STFT3D-CNN2D-V.ipynb

• STFT3D-CNN2D-V-DataAug-Explorar.ipynb

• STFT3D-CNN2D-V-Resize.ipynb

• STFT3D-ConvLSTM2D-V-Resize.ipynb

• STFT3D-RNNs-2C-V.ipynb

• STFT3D-RNNs-H.ipynb

• STFT3D-RNNs-Ventaneo.ipynb

• STFT3D-Ventaneo.ipynb

• STFT3D-Ventaneo-DataAugmentation.ipynb

• STFT4D.ipynb

• STFT4D-2C.ipynb

• STFT4D-2C-Resize.ipynb

• STFT4D-Resize.ipynb

• STFT-CNN2D-DA_2.ipynb

• Transfer-Learning.ipynb

• 2C = 2 canales: Los canales son C3 y C4

• V/H = Concatenacion vertical u horizontal

• Resize = resize de la imagen

• Explorar = Impresion de ejemplos

• DataAugmentation = width_shift_range

• Ventaneo = Ventanas de 2 segundos superpuestas al 95%

• x_2 = 2 segundos de señal

Escalogramas-CWT

• Escalogramas con la transformada continua wavelet, modulo de la wavelet de Morlet compleja (cmor3-3), 128 escalas

• CWT-CNN.ipynb

• CWT-CNN-CV.ipynb

• CWT-CNN-GS.ipynb

• CWT-CNN-Predecir.ipynb

• CWT-CNN-LSTM.ipynb

• CWT-CNN-LSTM-CV.ipynb

• CWT-CNN-LSTM-GS.ipynb

• CWT-CNN-LSTN-Predecir.ipynb

• CWT-CNN-Explorar.ipynb

• CWT3D-CNN2D-2C-V.ipynb

• CWT3D-CNN2D-V.ipynb

• CWT3D-RNNs-2C-V.ipynb

• CWT3D-Ventaneo.ipynb

• CWT3D-Ventaneo-DataAugmentation.ipynb

• CWT4D.ipynb

• CWT4D-2C.ipynb

• CWT-CNN2D-DA_2.ipynb

• Fine-Tuning.ipynb

• Transfer-Learning.ipynb

• 2C = 2 canales: Los canales son C3 y C4

• V = Concatenacion vertical

• Explorar = Impresion de ejemplos

• DataAugmentation = width_shift_range

• Ventaneo = Ventanas de 2 segundos superpuestas al 95%

• x_2 = 2 segundos de señal

Notas

• Al importar los datos header=None porque no hay encabezado
• Las etiquetas tienen valores de 1 a 4 por lo tanto en Keras se tienen que codificar las etiquetas u otra opción es procesar las etiquetas para que los valores sean del 0 al 3

Mas información

https://github.com/EdgarMoyete/EEG-Classification-with-CNN

deep-learning-bci-iv-2a's Issues

Dataset

Can you provide the csv of the dataset?

Consult BCI 2a CSV (288,22000) data information

Hello author, through the CSV data of BCI 2A you shared, I realized a neural network and got good classification results. However, when I explained the network to my teacher, the teacher pointed out a point that I had ignored: the information problem carried by the data itself. The data of 9 people in CSV shared are all the data volume of (288,22000). Here, I would like to ask you: 288 is understandable. There are four experimental paradigms recorded 72 times each, and 288 times in total. How does 22,000 understand that? What channels of data is he recording? Is it the data of BCI 3S or 1+ 3S with 1s extra rest time?If you have time to answer, it will be very helpful to me! Anyway, thank you again!

About the preprocess?

Hello, I would like to ask what are the preprocessing steps of the data? The data format is ndarray: (288 ~ 22 ~ 1000). Does it include artifact removal, such as ICA? Is a rereference used? Is baseline correction done?

Email about consulting csv

Hello sir, I sent an email to your email on 2022.08.10, my email is [email protected] If you see it, please check it out during your busy schedule, thank you very much!

how can i get the model 'CWT09.hdf5'?

Hello! Your job is fantastic!I wonder where some models for example 'CWT09.hdf5' are?It was used in the file Escalogramas-CWT/CWT-CNN-Predecir.ipynb.