An Efficient Radar-Based Gesture Recognition Method Using Enhanced GMM and Hybrid SNN

This article proposes an energy-efficient and high-accuracy gesture recognition framework to address the challenges of high computational complexity and interference susceptibility in conventional radar-based gesture recognition methods. First, an enhanced Gaussian mixture model (GMM) with an optimized learning rate is introduced to improve anti-interference performance by exploiting spatial and velocity differences between gesture signals and target-like interference. Furthermore, a novel spiking neural network (SNN) architecture is proposed, combining a 2-D convolutional neural-network (2D-CNN) for spatial feature extraction with a long short-term memory (LSTM) network for capturing, long-term temporal dependencies. This hybrid architecture effectively integrates short-term and long-term temporal dynamics to enhance recognition accuracy. Additionally, spike-timing-dependent plasticity (STDP) is incorporated to address the non-differentiability of spike-based data, thereby improving the network’s feature learning capabilities. To evaluate the proposed approach, a radar-based gesture dataset comprising seven gesture categories was constructed using a 60-GHz frequency-modulated continuous wave (FMCW) radar system. Experimental results demonstrate a recognition accuracy of 99.28%, alongside computational complexity and power consumption have better performance than the existing competitive methods, suiting power and resource-constrained environments.