Gesture Recognition of Surface Electromyography Signal Based on Multi-source Domain Transfer Learning

In the field of gesture recognition, solutions based on surface EMG signals have significant advantages. However, the individual differences among users result in low accuracy when recognizing new user data directly. In practical applications, extracting the common features of surface EMG signals from different users poses a serious challenge for the model. To address this issue, this paper proposes a framework called Multi-Source Integration based Transfer Learning. This framework aims to improve the transfer of knowledge from source users to target users by placing them in their respective feature spaces. The experimental results show that the method achieves an average accuracy of 72.15% and 76.62% respectively on the NinaproDB5 datasets. This improvement in classification accuracy applies to both within-subjects and between subjects cases, enhancing the model’s clinical applicability in practical scenarios.