Related Experiment Video
Updated: Aug 20, 2025
08:15
Capturing Dynamic Finger Gesturing with High-resolution Surface Electromyography and Computer Vision
Published on: March 28, 2025
704
The spike gating flow: A hierarchical structure-based spiking neural network for online gesture recognition
Zihao Zhao1,2, Yanhong Wang1,2, Qiaosha Zou1
1School of Microelectronics, Fudan University, Shanghai, China.
Frontiers in Neuroscience
|November 21, 2022
Summary
This study introduces Spiking Gating Flow (SGF), a brain-inspired system for efficient online action recognition. SGF achieves high accuracy with minimal data and training, outperforming traditional deep learning methods.
Area of Science:
- Artificial Intelligence
- Computer Vision
- Neuroscience
Background:
- Current deep learning (DL) methods for action recognition face challenges in computational cost and learning efficiency.
- Emerging applications in robotic vision and autonomous vehicles require more efficient AI solutions.
Purpose of the Study:
- To develop a novel, brain-inspired Spiking Neural Network (SNN) system for efficient online action learning.
- To address the limitations of traditional DL in terms of computational cost and data requirements.
Main Methods:
- Developed a hierarchical system of Spiking Gating Flow (SGF) units for online action learning.
- Each SGF unit comprises feature extraction, event-driven, and histogram-based training layers.
- Utilized a dynamic vision sensor (DVS) for benchmark gesture classification.
Main Results:
- Achieved 87.5% accuracy in DVS gesture classification, comparable to DL methods.
- Demonstrated a low training/inference data ratio of 1.5:1 and required only a single training epoch.
- Attained the highest accuracy among non-backpropagation-based SNNs to date.
Conclusions:
- The developed SGF network enables few-shot learning (FSL) through a hierarchical structure incorporating prior knowledge.
- SNNs are effective for content-based global dynamic feature detection in action recognition.
