A CNN-LSTM model for six human ankle movements classification on different loads

Abstract

This study aims to address three problems in current studies in decoding the ankle movement intention for robot-assisted bilateral rehabilitation using surface electromyogram (sEMG) signals: (1) only up to four ankle movements could be identified while six ankle movements should be classified to provide better training; (2) feeding the raw sEMG signals directly into the neural network leads to high computational cost; and (3) load variation has large influence on classification accuracy. To achieve this, a convolutional neural network (CNN)-long short-term memory (LSTM) model, a time-domain feature selection method of the sEMG, and a two-step method are proposed. For the first time, the Boruta algorithm is used to select time-domain features of sEMG. The selected features, rather than raw sEMG signals are fed into the CNN-LSTM model. Hence, the number of model's parameters is reduced from 331,938 to 155,042, by half. Experiments are conducted to validate the proposed method. The results show that our method could classify six ankle movements with relatively good accuracy (95.73%). The accuracy of CNN-LSTM, CNN, and LSTM models with sEMG features as input are all higher than that of corresponding models with raw sEMG as input. The overall accuracy is improved from 73.23% to 93.50% using our two-step method for identifying the ankle movements with different loads. Our proposed CNN-LSTM model have the highest accuracy for ankle movements classification compared with CNN, LSTM, and Support Vector Machine (SVM).

Keywords: CNN; LSTM; SEMG signal; ankle movement classification; load variation.

Figure 1

Framework of the ankle movement classification method.

Figure 2

EMG electrodes positions.

Figure 3

Test benches. (A) Experimental setup. (B) Test bench of DF, PF, IV, and EV, the moving platform could rotate around the x-axis and y-axis. (C) Test bench of IR and ER, the moving platform could rotate around the z-axis.

Figure 4

Boruta algorithm procedures.

Figure 5

Window splitting method.

Figure 6

Experimental procedure.

Figure 7

Screenshot of the guidance interface of the experiment.

Figure 8

Result of the Boruta algorithm: Z-Score was used to represent the feature importance. The features suggested to be excluded are marked in red, features more related to the dependent variables are marked in green, and the shadow features are marked in blue.

Figure 9

Confusion matrix of classification results for all subjects under every single load using a one-step method. Each element represents the accuracy of all subjects for corresponding training and testing loads. Darker color indicates higher accuracy. The main diagonal elements represent the accuracies where the load of training and testing sets are the same. The off-diagonal elements represent the accuracies where the load of training and the testing sets are different.

Figure 10

Classification accuracy for six movements using different models. (A) CNN-LSTM model with sEMG features and raw sEMG as the inputs. (B) CNN model with sEMG features and raw sEMG as the inputs. (C) LSTM model with sEMG features and raw sEMG as the inputs. (D) Classification accuracy for six movements using different network architectures with feature extraction (*P-value < 0.05, **P-value < 0.01, and ***P-value < 0.001).

Figure 11

Distribution of features. (A) CNN-LSTM model input with sEMG features as the input. (B) CNN-LSTM model input with raw sEMG as the input.