Research on sEMG Feature Generation and Classification Performance Based on EBGAN

Surface electromyography signal (sEMG) recognition technology requires a large number of samples to ensure the accuracy training results. However, sEMG signals generally have problems small amount data, complicated acquisition process and environmental influence, which hinders improvement classification. In order improve classification, an feature generation method based on energy generative adversarial network (EBGAN) is proposed in this paper for first time. The concept introduced into discriminant instead traditional binary judgment, distribution real EMG dataset learned captured by multiple fully connected layers, with similar data being generated. experimental results show that, compared other types GAN networks, achieves maximum mean discrepancy comparison that original data. using different typical classification models augmentation can effectively models, increase range 1~5%.