EMG-Based Gesture Classification Using Convolutional Neural Networks

This study proposes a surface electromyography (sEMG) gesture classification method based on convolutional neural network (CNN). The gesture data of 30 subjects were collected experimentally. After preprocessing such as filtering and effective signal segment extraction, a one-dimensional CNN model was designed for automatic feature extraction and classification. The experimental results show that when the sliding window length is 300 and the amplitude multiple is 1.5, the classification accuracy of the model on the test set is 92.58%-98.63%. Compared with LSTM and SVM models, CNN performs better in 4-category, 6-category and 10-category gesture classification tasks, especially in complex gesture classification. The study verifies the effectiveness of CNN in sEMG signal classification and provides a direction for future optimization, such as further exploring the impact of network architecture and parameter configuration on performance to improve the ability of complex gesture recognition.