Spiking-Timing-Dependent Plasticity Convolutional Spiking Neural Network for Efficient Radar-Based Gesture Recognition

• Published in: 2023 International Conference on Image Processing, Computer Vision and Machine Learning (ICICML) pp. 617 - 620
• Main Authors: Wu, Yifan, Wu, Li, Xiao, Zelong, Hu, Taiyang
• Format: Conference Proceeding
• Language: English
• Published: IEEE 03.11.2023
• Subjects: component; Computational complexity; Computer architecture; Convolution; Gesture recognition; Power demand; SNN; STDP; Training; Unsupervised learning
• DOI: 10.1109/ICICML60161.2023.10424838

An innovative architecture, termed Spike-Timing-Dependent Plasticity Convolutional Spiking Neural Network (STDP-CSNN), is proposed for efficient radar-based gesture recognition in this paper. Radar range-Doppler image data are encoded into spike sequences with CSNN, and the leaky integrate-and-fire (LIF) model is employed as a neuron in the network nodes. This design enables efficient accumulation and transmission of spike signals, resulting in a significant reduction in network power consumption. Furthermore, the unsupervised learning algorithm of STDP is employed to facilitate feature extraction in the established CSNN, ensuring low computational complexity. The experimental results demonstrate that the proposed STDP-CSNN architecture achieves an impressive recognition accuracy of 92.72%, while concurrently addressing the crucial requirements of low power consumption and computational simplicity.