Open access
Sep 2026
Self-Supervised IMU-Based Human Activity Recognition with Deep Spatio-Temporal Feature Extraction and Adaptive Feature Fusion
The proposed SSL framework employs the Transformer and Depthwise Separable Convolution to jointly capture global temporal dependencies and local spatial features, which are adaptively fused into discriminative spatio-temporal representations for downstream HAR.
Qian Yang, Ying-Long Huang, Jin Han et al.
· Processes · 0 citations