Real-Time Hand Rehabilitation: FPGA-Accelerated Neural Network for Muscle Activity Classification

Detecting hand movement intention based on muscle activity is crucial for various assistive and rehabilitation applications. This paper presents an FPGA-based neural network designed to classify muscle activity as either stressed (movement intention) or relaxed (no movement intention). The neural network consists of a three-layer architecture (input, hidden, and output layers) and is implemented using High-Level Synthesis (HLS) for efficient hardware acceleration. The FPGA-based system processes muscle sensor data in real-time, offering improved responsiveness compared to traditional microcontroller-based implementations. To validate the effectiveness of the proposed method, the system is integrated into a wearable hand movement assistance prototype, where classified muscle states are used to trigger controlled hand motions. The FPGA implementation achieves low latency and power-efficient operation, making it suitable for real-time applications. Experimental results demonstrate that the hardware-accelerated neural network efficiently classifies muscle activity while maintaining a compact design. This work highlights the potential of FPGA-based neural networks in biomedical signal processing, offering a fast and efficient alternative to conventional microcontroller-based classification methods. Future work will focus on optimizing hardware resource utilization, improving classification accuracy, and extending the system for broader rehabilitation and prosthetic applications.