AUTHOR=Chen Zihan , Qian Yaojia , Wang Yuxi , Fang Yinfeng 

TITLE=Deep Convolutional Generative Adversarial Network-Based EMG Data Enhancement for Hand Motion Classification

JOURNAL=Frontiers in Bioengineering and Biotechnology

VOLUME=Volume 10 - 2022

YEAR=2022

URL=https://www.frontiersin.org/journals/bioengineering-and-biotechnology/articles/10.3389/fbioe.2022.909653

DOI=10.3389/fbioe.2022.909653

ISSN=2296-4185

ABSTRACT=The acquisition of bio-signal from human body requires strict experimental setup and ethic approvements, which leads to limited data for the training of classifiers in the era of big data. It will change the situation if synthetic data can be generated based on real data. This paper proposes such a kind of multiple channel electromyography (EMG) data enhancement method using deep convolutional generative adversarial network (DCGAN). The generation procedure is as follows. Firstly, the multiple channels of EMG signals within sliding windows are converted to grayscale image through matrix transformation, normalization, and histogram equalization. Secondly, the grayscale images of each class are used to train the DCGAN, so that synthetic grayscale images for each class can be generated with the input of random noises. To evaluate whether the synthetic data owns the similarity and diversity with the real data, the classification accuracy index is adopted in this paper. A public EMG dataset (i.e. ISRMyo-I) for hand motion recognition is utilized to prove the usability of the proposed method. Experimental results show that adding synthetic data into the training data does not influence the classification performance very much, indicating the similarity between real data and synthetic data. Besides, it is also found that with additional synthetic data for training, the average accuracy (five classes) is slightly increased by 1% and 2% for support vector machine (SVM) and random forest (RF), respectively. Although the improvement is not statistically significant, it implies that the generated data by DCGAN own its new characteristics, and it is possible to enrich the diversity of training dataset.  In addition, cross-validation analysis shows that the synthetic samples have large inter-class distance, reflected by higher cross-validation accuracy for pure synthetic sample classification.  Furthermore, this paper also demonstrates that histogram equalization can significantly improve the performance of EMG based hand motion recognition.