Temporal and Modality Awareness-Based Lightweight Residual Network with Attention Mechanism for Human Activity Recognition Using a Lower-Limb Exoskeleton Robot

This study proposes a novel lightweight residual network for recognizing diverse locomotion activities across varying terrains using multimodal sensing data. The model adopts an asymmetric convolutional architecture composed of depthwise and pointwise layers to efficiently capture both temporal and modality-specific features while significantly reducing the number of trainable parameters. A channel-attention block is further integrated to emphasize salient fused features. Evaluations conducted using the Walking Assist Wearable Robot Motion dataset, which contains kinematic and postural signals from 500 adults wearing a lower-limb exoskeleton, demonstrated that the proposed model achieves an accuracy of 98.23% and a macro F1 score of 98.21%. These results outperform those of four hybrid deep learning baselines while reducing the parameter count by 5.5 to 12.9 times. The findings suggest that temporal features alone are insufficient for robust human activity recognition; modality-level features play a critical and complementary role. To assess the generalizability of the proposed method, additional experiments were conducted on four public benchmark datasets, UCI-HAR, HAPT, PAMAP2, and WISDM, under varying batch size conditions. Compared with hybrid baselines, the model consistently achieved competitive or superior performance, with macro F1 scores of 0.9627 ± 0.0046, 0.776 ± 0.0138, 0.8937 ± 0.0121, and 0.9736 ± 0.0025, respectively. These results confirm the robustness and adaptability of the proposed architecture across a wide range of real-world scenarios.