Hand gesture recognition based on surface electromyography(sEMG)signals is widely used in the field of human-computer interaction

accurately and quickly recognizing gesture actions can provide a better user experience.However

the overall recognition accuracy is low in multitask scenarios due to individual differences.An sEMG gesture recognition method based on densely connected convolutional network(DenseNet) and multi-domain feature fusion is proposed.Firstly

time-domain and frequency-domain features are extracted from sEMG signals to form a feature set

which is fused with the original signals as network inputs to enhance the representation ability of input data.Secondly

a DenseNet network fusing squeeze-and-excitation(SE) attention and multi-scale dilated convolution is used for feature extraction and classification recognition.Experimental results show that the overall recognition accuracy of gesture recognition reaches 88.06% on the NinaPro DB2 dataset

and the recognition performance is improved in both overall and classified movements.