Cancelable HD-SEMG Biometric Identification via Deep Feature Learning.

Area of Science:

• Biometrics and Human-Computer Interaction
• Signal Processing and Machine Learning

Background:

• Conventional biometrics (face, fingerprint, iris) are susceptible to imitation and circumvention.
• Secure and cancelable biometric modalities are crucial for personal identification, particularly in smart healthcare.
• Existing methods lack robust security and customizability for evolving identification needs.

Purpose of the Study:

• To develop a novel person identification model utilizing high-density surface electromyography (HD-sEMG) as a secure biometric trait.
• To create cancelable HD-sEMG biometric templates customizable through user-specific finger isometric contractions.
• To evaluate the identification accuracy and efficiency of a deep feature learning approach for HD-sEMG-based biometrics.

Main Methods:

• Development of a person identification model using high-density surface electromyography (HD-sEMG) signals.
• Implementation of cancelable biometric templates through user-controlled finger isometric contractions.
• Application of convolutional neural networks (CNNs) for deep feature learning and user identification from HD-sEMG data.

Main Results:

• The proposed HD-sEMG biometric model achieved a rank-1 identification accuracy of 87.23% and an equal error rate of 4.66% for 44 identities.
• Cross-day identification accuracy surpassed previous methods, demonstrating robust performance.
• Usability and efficiency investigations indicated significant potential for practical applications in secure identification.

Conclusions:

• HD-sEMG offers a secure and cancelable biometric modality with high identification accuracy.
• The deep learning approach effectively captures user-specific patterns for reliable person identification.
• The developed model shows promise for enhancing security in smart healthcare and other applications requiring robust personal identification.