Application of the LSTM-SNN Hybrid Temporal Modeling Approach in Complex Gait Activity Recognition

Gait activity recognition plays a crucial role in fields such as rehabilitation medicine, health monitoring, and smart devices. To address the challenges in complex gait activity recognition, such as difficulties in modeling long time-series, detecting activity transitions, and integrating multimodal data, this paper proposes a multimodal temporal modeling approach that combines Long Short-Term Memory (LSTM) networks and Spiking Neural Networks (SNN). This approach leverages LSTM to extract global dependency features from long time-series and capture long-term patterns during activity transitions. Meanwhile, SNN simulates the pulse dynamics of biological neural systems, enhancing the ability to perceive short-term features. To further improve recognition performance, the model integrates multimodal data from inertial sensors (accelerometers and gyroscopes) and electromyography (EMG) signals, generating a unified feature representation, and combines LSTM and SNN for fine-grained classification.