A Lightweight CNN Approach for Hand Gesture Recognition via GAF Encoding of A-Mode Ultrasound Signals.

Area of Science:

• Biomedical Engineering
• Computer Science
• Artificial Intelligence

Background:

• Hand gesture recognition (HGR) is crucial for human-computer interaction.
• Existing methods often require complex models or lack efficiency.
• A-mode ultrasound signals offer a novel data source for HGR.

Purpose of the Study:

• To develop an efficient and accurate HGR system using A-mode ultrasound signals.
• To introduce a novel lightweight, parameter-free attention convolutional neural network (LPA-CNN).
• To leverage Gramian Angular Field (GAF) transformation for signal processing.

Main Methods:

• 1D A-mode ultrasound signals from forearm muscles were transformed into 2D images using GAF.
• A novel LPA-CNN architecture was designed, incorporating convolution-pooling, attention mechanisms, inverted residual blocks, and classification layers.
• Comparative experiments were conducted against GoogLeNet and MobileNet.

Main Results:

• The proposed LPA-CNN achieved a classification accuracy of 0.98 ± 0.02.
• LPA-CNN demonstrated a smaller model size compared to GoogLeNet and MobileNet.
• The GAF transformation proved more sensitive for HGR than MTF and RP.

Conclusions:

• The integration of GAF transformation and LPA-CNN provides an efficient and high-accuracy approach for HGR.
• This method offers a new technological avenue for HGR utilizing ultrasonic signals.
• The developed system enhances human-computer interaction through precise gesture recognition.