Hybrid Attention with CNN-BiLSTM and CBAM for Efficient Wearable Activity Recognition
Sakorn Mekruksavanich, Ponnipa Jantawong, Wikanda Phaphan, Anuchit Jitpattanakul · 2024
The use of widespread sensors for automatic activity recognition has become a dynamic area of research due to its broad applications in healthcare, fitness monitoring, and assisted living. To efficiently identify daily and sports activities from sensor data, this study suggests a hybrid deep learning model that integrates convolutional neural networks (CNN), bidirectional long short-term memory (BiLSTM) networks, and convolutional block attention modules (CBAM). The model starts by extracting spatial features from the sensor data using CNN layers. These spatial features are then employed to offer temporal context information to the BiLSTM network. To conclude, the CBAM attention mechanism is applied to guide the model’s focus to the most informative segments of the BiLSTM feature maps. This approach enables the accurate recognition of patterns in both complex and daily sports activities. The effectiveness of the proposed hybrid model is evaluated using the UCI-DSA, a publicly available benchmark dataset that includes sports and daily activities collected through wearable motion sensors. In contrast to existing approaches, the outcomes reveal that a unified framework involving CNN, BiLSTM, and CBAM achieves outstanding performance in activity recognition, exceeding $99 \%$ in both accuracy and F1-score while maintaining computational efficiency.