Gait phase detection based on inception-self-attention

Gait analysis is an important research direction in motion monitoring and lower limb exoskeleton. Based on real IMU data and ground reaction force data, this paper proposes a deep learning model that combines Inception and Self-Attention mechanisms for gait phase classification. The experiment uses multi-time step input to perform classification tasks under 4-phase, 5-phase, and 6-phase gait division conditions and comprehensively analyzes the impact of model parameters on performance. The experimental results show that the classification accuracy of the proposed method under different divisions reaches 97.85%, 96.77%, and 96.37% respectively, reflecting the model’s adaptability to different gait fine-grained classification tasks. This study provides effective support for the application of gait recognition technology in the development of smart wearable devices and motion monitoring and also lays a theoretical and practical foundation for in-depth research on gait phase classification.