Simultaneous Recognition of Finger Flexion, Angle, and Force Based on a Wearable High-Density Neuromuscular Interface for Real-Time Anthropomorphic Prosthetics Control

In the field of prosthetics hand control, finger movements offer greater dexterity and operation precision than conventional hand gesture and wrist gesture, enabling fine-grained human-computer interaction tasks, such as traditional Chinese medicine sphygmopalpation and laboratory hazardous reagent operation. These tasks involve finger flexion, flexion angle and fingertip force. However, few studies have simultaneously recognized these motion information, and applied them to real-time prosthetic hand control. In this paper, we present a wearable high-density surface electromyography (HD-sEMG)-based system for simultaneous recognition of finger flexion, flexion angle and fingertip force. The system incorporates a flexible and stretchable electrode array with a portable wireless acquisition device to record high-resolution and high-sampling-rate sEMG data. Then, a convolutional neural network processes three-dimensional (3D) sEMG data was introduced to decode finger flexion, flexion angle and fingertip force. Experimental results demonstrate that utilizing the 3D sEMG data improves classification accuracy by over 7% compared to conventional two-dimensional (2D) sEMG data. Furthermore, we validated the real-time performance of the developed system by controlling a prosthetics hand to perform finger flexions with different flexion angles and fingertip forces. As a practical application, translating the recognition results into real-time prosthetics control successfully demonstrated the system's capability to replicate diverse sphygmopalpation gestures, highlighting the system's potential for clinical diagnostics and other high-precision applications.