Deep Learning-Based Biometric Payment System with Dual-Tap Fingerprint Authentication

Mobile payment fraud and identity theft pose critical threats to the rapidly expanding digital financial ecosystem, projected to exceed $15 trillion in transaction volume by 2025. Traditional authentication methods including passwords and one-time passwords (OTPs) remain vulnerable to phishing, credential stuffing, and replay attacks. This paper proposes a novel three-layer biometric payment framework integrating: (1) a dual deep learning architecture employing a CNN-based liveness detection model and a Siamese/Triplet network for fingerprint embedding, (2) a Dual-Tap Fingerprint Authentication (DTFA) protocol that introduces temporal behavioral biometrics through configurable two-stage fingerprint capture, and (3) a Dynamic One-Time Transaction (DOT) Code layer using SHA-256 cryptographic binding. Evaluated on a custom dataset of 5,000 participants encompassing 50,000 genuine fingerprint samples and 15,000 presentation attacks, the proposed DTFA-DOT system achieves 99.47% authentication accuracy with an Equal Error Rate (EER) of 0.085%, a False Acceptance Rate (FAR) of 0.02%, and 0% success rate against replay, man-in-the-middle, and transaction manipulation attacks. User acceptance studies with 500 participants yielded a 4.6/5.0 satisfaction rating and 8.2-second average transaction time. These results demonstrate a new standard for secure, usable mobile banking authentication.