Deep Forest Model Combined with Neural Networks for Finger Joint Continuous Angle Decoding

Many people lose their hand function due to stroke, traffic accidents, and amputation. This paper proposed a new method that can decode hand joint angles from the upper limb’s surface electromyography (sEMG). It can be used for the next generation of prosthetic hands, and rehabilitation exoskeletons which will be beneficial for patients and amputees. We simultaneously collect hand joints’ angles and sEMG signals by VICON and Noraxon systems. We combine a deep forest algorithm with an artificial neural network to design a comprehensive decoding model. Compared with the Gaussian Process model, its average correlation coefficient has improved by 42%, reaching 0.844. This method shows great potential in prosthetic hand and exoskeleton control. We have also carried out online experiments in which the online experiments achieved a completion rate of more than 90%, as well as a low latency, which is ideal for realistic online prosthetic control scenario applications.