Deep Residual Network for Smartwatch-Based User Identification through Complex Hand Movements

Area of Science:

• Computer Science
• Human-Computer Interaction
• Biometrics

Background:

• Wearable technology is increasingly used in daily life, necessitating secure authentication.
• Current activity-based user identification methods are limited.
• Research on using complex hand movements for user identification is scarce.

Purpose of the Study:

• To investigate the effectiveness of complex hand movements for user identification using wearable technology.
• To develop and evaluate a deep learning model for hand movement-based authentication.

Main Methods:

• Utilized a one-dimensional residual network with squeeze-and-excitation (SE) configurations (1D-ResNet-SE).
• Analyzed hand movement data from smartwatch sensors.
• Employed deep learning for feature extraction and user identification.

Main Results:

• The 1D-ResNet-SE model demonstrated superior performance in user identification compared to other models.
• SE modules significantly enhanced the identification capabilities of the residual network.
• The deep learning approach effectively identified features from smartwatch sensor data.

Conclusions:

• Hand movement assessment using deep learning is a viable and effective method for smartwatch user identification.
• The proposed 1D-ResNet-SE model offers a promising solution for secure, non-intrusive authentication in wearable systems.