For patient with physical disability, rehabilitation robots can help them restore or augment limbs function, and someday may replace physiotherapist. For the study of rehabilitation robot, human motion intention recognition based on surface electromyography (sEMG) has been regarded as an outstanding way to realize the intuitive and multifunctional control of prosthesis and rehabilitation hands.

Recently, ZHAO Xingang and his research team at Shenyang Institute of Automation of Chinese Academy of Sciences have proposed an adaptive hybrid classifier for sEMG recognition under non-ideal conditions such as the outlier motion, muscle fatigue, and electrode shifts. The study was published in IEEE Transactions on Neural Systems and Rehabilitation Engineering.

The researchers have achieved recognition accuracy of 92 percent for ten common gestures, which is significantly higher than that of previous studies. This study is expected to promote the clinical applications of sEMG-based prosthesis and rehabilitation hands.

However, it has been found that there is a big gap between academic research and clinical application. One key factor is the experimental conditions for sEMG-based recognition in the laboratory are too simple and without daily interferences. To solve this problem, the research team proposed an adaptive hybrid classifier to deal with various interference problems in clinical applications such as new movements, muscle fatigue and electrode migration.

They first constructed an adaptive incremental hybrid classifier (AIHC) by combining one-class LDA in conjunction with two update schemes. The AIHC can reject outlier classes, and its recognition ability can incremental grow online by considering the rejected class as a new target class.

Besides, they designed an online evaluation factor to help AIHC self-update for adapting to the changes in sEMG characteristics.

Finally, an AIHC-based recognition strategy was developed to reduce the influence of multiple interferences caused by outlier motion, muscle fatigue, and electrode shifts. The obtained recognition results indicate that the AIHC can accommodate a large-scale deviation in sEMG characteristics and has significant advantages compared to previous studies.

Adaptive hybrid classifier for myoelectric recognition in non-ideal conditions (Image by ZHAO Xingang)