sEMG-Based Hand Gesture Recognition Using Binarized Neural Network.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Wearable Technology

Background:

• Human-machine interfaces (HMI) are crucial for convenient device control.
• Hand gesture recognition (HGR) systems offer intuitive interaction.
• Surface electromyography (sEMG) sensors are advantageous for HGR due to environmental independence and low data requirements.

Purpose of the Study:

• To develop a compact and efficient HGR system overcoming limitations of bulky multi-sensor setups and complex deep learning models.
• To implement a lightweight convolutional neural network (CNN) using a binarized neural network (BNN) approach.
• To integrate the system onto a field-programmable gate array (FPGA) for practical application.

Main Methods:

• Utilized a single dry-type sEMG sensor for data acquisition.
• Employed a binarized neural network (BNN), a lightweight CNN, for gesture classification.
• Converted raw sEMG data into spectrograms for time-frequency domain analysis.
• Implemented the system on a field-programmable gate array (FPGA).

Main Results:

• Achieved 95.4% classification accuracy for nine dynamic gestures.
• Demonstrated a low computation time of 14.1 ms.
• Reported a power consumption of 91.81 mW, indicating high efficiency.

Conclusions:

• The proposed HGR system effectively recognizes dynamic gestures with high accuracy using minimal hardware.
• The BNN-based approach on FPGA offers a practical solution for wearable HMI, balancing performance and resource constraints.
• This technology has the potential to enhance user interaction across various applications.