A Novel Spiking Neural Network for Wearable-Based Human Activity Recognition

Human activity recognition (HAR) tasks based on deep learning often suffer from class imbalance, which leads to model overfitting on high-frequency classes. Existing approaches typically require the introduction of complex modules, resulting in significant computational overhead. To address these limitations, we propose a lightweight spiking neural network (SNN) based on attention mechanism, specifically designed to achieve competitive performance in HAR. By introducing adaptive class weights for data sampling and a temporal attention module (TAM), along with the integration of leaky integrate-and-fire (LIF) neurons, our approach achieves state-of-the-art accuracy rates of 99.08% and 94.34% on the UCI-HAR and UniMiB SHAR datasets, respectively. Experimental results demonstrate that our model excels in recognition accuracy and class balance.