Ultrasonic Guided Wave Inversion Based on Deep Learning Restoration for Fingerprint Recognition.

Area of Science:

• Biometrics and Signal Processing
• Non-Destructive Testing
• Deep Learning Applications

Background:

• Fingerprint scanning is a reliable biometric authentication method.
• Accurate fingerprint reconstruction is crucial for security and identification.
• Existing methods face challenges in large domains and computational efficiency.

Purpose of the Study:

• To achieve accurate fingerprint reconstruction in large physical domains using ultrasonic guided waves.
• To enhance fingerprint minutiae detection and characterization.
• To improve the efficiency and accuracy of fingerprint reconstruction through deep learning.

Main Methods:

• Utilizing ultrasonic guided waves to monitor wavefield variations in plate-like structures.
• Employing fast inversion tomography (FIT) integrated with a deep learning convolutional neural network (DLCNN).
• Conducting parametric optimization and developing a specific DLCNN model for artifact removal in FIT reconstructions.

Main Results:

• Demonstrated accurate fingerprint reconstruction and quantitative characterization at any position.
• Achieved submillimeter fingerprint minutiae detection through parametric optimization.
• Significantly improved imaging quality, including increased resolution and reduced reconstruction errors.
• Reported exponential improvement in computational efficiency with reduced sensor numbers.

Conclusions:

• The FIT method based on DLCNN restoration substantially enhances fingerprint imaging quality and matching confidence.
• The proposed approach offers a computationally efficient solution for fingerprint reconstruction in large domains.
• This research advances biometric authentication through improved ultrasonic-based fingerprint analysis.