Pattern Recognition of EMG Signals by Machine Learning for the Control of a Manipulator Robot

. 2022 Apr 30;22(9):3424. doi: 10.3390/s22093424

Algorithm 1: Multilayer Perceptrón algorithm implemented for the EMG

p \leftarrow Input_vector

T \leftarrow Output_vector

3 /*** output vector T where the labeling value is stored by one-hot-encoding of each the classes***/

T \leftarrow [\{0, 0, 0, 1\}, \{0, 0, 2, 0\}, \{0, 3, 0, 0\}, \{4, 0, 0, 0\}]

scaler \leftarrow StandardScaler () . fit (P)

p \leftarrow scaler . transform (P)

7 /**Divide p into a test (

P_{test})

and a training set

(P_{train})

**/

one_hot_labels = to_categorial (T, num_classes \leftarrow 4)

P_train, P_test, T_train, T_test \leftarrow train_test_split (P, one_hot_labels, test_size \leftarrow 0.20, random_state \leftarrow 42)

10 /**Random Initialization**/

W \leftarrow 2 \times (random - 0.5) \times scale

epochs \leftarrow 3000

hiddenNodes \leftarrow 4

model \leftarrow Sequential ()

model . add (Dense (hiddenNodes, activation \leftarrow relu, input_\dim \leftarrow 4)

a [1] \leftarrow \max (0, n)

//ReLu activation function

model . add (Dense (4, activation \leftarrow' softmax'))

a [2] \leftarrow e^{n_{4}} / \sum_{1}^{5} e^{n_{4}}

//Softmax activation function

model . summary ()

loss \leftarrow categorical_crossentropy

21 /**Loss function (categorial cross entropy**/

L (y, \hat{y}) \leftarrow \frac{1}{N} \sum_{j = 1}^{M} \sum_{i = 1}^{N} (y_{ij} \log ({\hat{y}}_{ij}))

optimizer \leftarrow tf . keras . optimzers . Adam ()

W \leftarrow W - \frac{α m}{\sqrt{v + ϵ}}

model . compile (loss \leftarrow loss, optimizer \leftarrow optimizer, metrics \leftarrow [' accuracy'])

history \leftarrow model . fit (P_train, T_train, epochs \leftarrow epochs, vebose \leftarrow 1, validation_split \leftarrow 0.1)

test, test \leftarrow model . evaluate (P, t, verbose \leftarrow 1)

weights (model . layers, 3)

scaling (scaler, 3)

layers (model . layers)