A LIGHTWEIGHT CNN-LSTM HYBRID MODEL FOR ANKLE MOTION RECOGNITION AND CLASSIFICATION BASED sEMG

In recent years, research on decoding human movement intentions using surface electromyography (sEMG) has garnered increasing attention in the field of human-computer interaction control.To enhance the accuracy of controllers, the adoption of effective signal processing methods and prediction models has become necessary.Based on this, a lightweight deep learning algorithm-Convolutional Neural Network (CNN)-Long Short-Term Memory (LSTM) hybrid model-is proposed in this paper, which aims to address three key issues in ankle movement recognition and classification:(1) Most of the existing methods can only recognize four types of ankle joint movements; and to achieve more efficient training and application, the classification of six types of ankle joint movements is necessary.(2) Currently, the CNN-LSTM model based on sEMG not only has high structural complexity but also has unsatisfactory classification accuracy.(3) The existing models have a slow convergence speed and a low utilization rate of computing resources.The lightweight CNN-LSTM hybrid model proposed in this paper can predict and classify six types of ankle movements.After 17 iterations, the model converged to the target state, achieving a classification accuracy of 99.46%.The hybrid model comprises a CNN layer, an LSTM layer, and a fully connected layer.Among them, the CNN layer can automatically extract features from sEMG signals, while the LSTM layer can capture long-term temporal dependencies in sEMG data.By comparing the output results of the CNN and LSTM models with those of the hybrid model, the experimental results show that the hybrid model outperforms both the individual CNN model and the LSTM model in terms of performance.In addition, the proposed CNN-LSTM hybrid model was compared in this paper with other commonly used classification algorithms, including RNN-GRU, RF, SVM, and LDA.Experimental results demonstrate that the CNN-LSTM model achieves the best classification accuracy.