Sonomyography-Based Proportional Control For Isokinetic Movements: A Comparative Analysis Of Features And Algorithms

Sonomyography is a technique that uses ultrasound imaging to measure muscle activity. Unlike traditional methods, sonomyography allows for real-time imaging of deeper muscles. Recent research has focused on developing wearable ultrasound devices to enable real-time muscle activity sensing, specifically for applications in proportional control systems. By tracking muscle movement intensity and dynamics, wearable sonomyography systems enable precise control in real-time applications, such as target tracking and robotic prosthetics, improving the functionality and user-friendliness of biomechatronic interfaces. In this paper, we explore the proportional control for target tracking using the sonomyography (SMG) signals, captured through a sensor array. We evaluated three feature extraction methods for comparison: peak location-based features, amplitude histogram features, and curve-fitting features. These features were used to train a regression model, which we evaluated for its effectiveness in controlling real-time gestures. The results demonstrate that the amplitude histogram feature set achieved the highest accuracy of $88.72 \pm 3.27 \%$ when trained with an ensemble regression model during the target tracking proportional control task using the SMG signal.