A multi-scale feature extraction fusion model for human activity recognition.

Area of Science:

• Computer Science
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Deep learning (DL) methods excel in Human Activity Recognition (HAR) due to automatic feature extraction.
• However, DL models often demand significant computational resources, limiting their use on resource-constrained devices like smartphones.
• There is a need for accurate HAR models that are computationally efficient.

Purpose of the Study:

• To develop a Human Activity Recognition (HAR) model that reduces computational cost and model size while maintaining high recognition accuracy.
• To address the challenges of deploying DL-based HAR on mobile devices.

Main Methods:

• Proposed a multi-scale feature extraction fusion model combining Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU).
• Utilized different convolutional kernel sizes with GRU for extracting local features and long-term dependencies.
• Employed separable convolutions instead of classical convolutions to decrease model parameters.

Main Results:

• Achieved high accuracy rates: 97.18% on WISDM, 96.71% on UCI-HAR, and 96.28% on PAMAP2 datasets.
• Demonstrated lower computational resource usage compared to existing methods.
• Successfully reduced model parameters while improving recognition accuracy.

Conclusions:

• The proposed CNN-GRU model effectively extracts rich features for HAR with improved efficiency.
• The model is suitable for deployment on devices with limited memory and computational power.
• This approach offers a promising solution for efficient and accurate Human Activity Recognition.