As part of studying deep learning and Keras, I am trying to implement the following line of pipelines:
The idea is this:
- EEG data input sections (6000x1 is what I will use for now)
- run it through 20 1D filters (200x1)
- make the maximum combination at the output of each of these filters with pool 20, step 10 (as a result, 20 data points 578x1 are obtained)
- "drain" it into a 578x20 matrix
- run this through a two-dimensional convolution with a kernel size of 30x20
- Maxpool again, with pool (10.1), increments (2.1)
- two consecutive fully connected levels
- 5-class softmax function.
My code is:
model = Sequential()
model.add(Conv1D(input_shape=(6000,1),kernel_size=200,strides=1,
activation='relu',filters=20,name='C1'))
model.add(MaxPooling1D(pool_size=20, strides=10,padding='valid',name='P1'))
model.add(Reshape(( 579, 20,1),name='S1'))
model.add(Conv2D(filters=400,kernel_size=(30,20),strides=(1,1),
activation='relu',name='C2'))
model.add(MaxPooling2D(pool_size=(10,1),strides=(2,1),padding='valid',name='P2'))
#model.add(Reshape((271*400,1,1),name='S2'))
model.add(Dense(500,activation='relu',name='F1'))
model.add(Dense(500,activation='relu',name='F2'))
model.add(Dense(5,activation='relu',name='output'))
model.add(Activation(activation='softmax',name='softmax'))
model.summary()
The result of this:
Layer (type) Output Shape Param #
=================================================================
C1 (Conv1D) (None, 5801, 20) 4020
_________________________________________________________________
P1 (MaxPooling1D) (None, 579, 20) 0
_________________________________________________________________
S1 (Reshape) (None, 579, 20, 1) 0
_________________________________________________________________
C2 (Conv2D) (None, 550, 1, 400) 240400
_________________________________________________________________
P2 (MaxPooling2D) (None, 271, 1, 400) 0
_________________________________________________________________
F1 (Dense) (None, 271, 1, 500) 200500
_________________________________________________________________
F2 (Dense) (None, 271, 1, 500) 250500
_________________________________________________________________
output (Dense) (None, 271, 1, 5) 2505
_________________________________________________________________
softmax (Activation) (None, 271, 1, 5) 0
=================================================================
Total params: 697,925.0
Trainable params: 697,925.0
Non-trainable params: 0.0
_________________________________________________________________
. , F1 500x1 (500 ), , ? P2 F1? "model.add(Reshape ((271 * 400,1,1), name= 'S2'))" , P2 .
"image_data_format": "channels_last" keras.json, , - - col-col-channel?
, .