Electrode Shift Fast Adaptive Correction for Improving Myoelectric Control Interface Performance

The emergence of wearable myoelectric armbands has greatly enhanced the convenience and efficiency of users in utilizing myoelectric gesture control interfaces. However, in practical applications, electrode shift can significantly degrade the performance of recognition models. To improve this situation, this study first proposes an electrode shift fast correction (ESFC) method for electrode shift after rewearing the device. Subsequently, a recognition model online update framework based on unsupervised subdomain adaptation is proposed to handle electrode shift during usage. In addition, the initial recognition model is trained using augmented data based on potential shift positions. Offline experiments were conducted using ten subjects from a public dataset. The results demonstrated that the ESFC method can correct any electrode shift angle to be within the range of -22.5°to 22.5° by performing a 1-s correction gesture. The average accuracy of the initial recognition model reached 87.09%, which was better than the comparison methods. After one update of the recognition model, the accuracy further improved to 90.20%. Six subjects were involved in the online experiments, and the results showed that the proposed online updating framework effectively improved the robustness of the recognition model to electrode shift during usage. The accuracy at the three stages reached 90.72%, 87.80%, and 87.83%, which was significantly higher than the methods that did not undergo online updating. The electrode shift scenario in this study is highly similar to real-world situations, indicating that the proposed method has great potential for practical applications.