Deep Learning and Physical Unclonable Functions for Secure Authentication in Smart Environments

User and device authentication is essential in smart environments to guarantee secure access to delicate information and systems. Traditional authentication methods, such as passwords and biometrics, often have significant vulnerabilities, including susceptibility to brute-force attacks and replication. This paper proposes an innovative authentication mechanism that leverages the strengths of Physical Unclonable Functions (PUFs) for generating unique device/user signatures and Siamese Neural Networks (SNNs) to accurately compare these signatures with certified ones. PUFs exploit the intrinsic manufacturing variations of hardware components to provide a unique, unclonable identifier for each user or device. In particular, optical PUFs use the inherent randomness and microscopic imperfections generated during the fabrication process in optical materials or structures. SNNs, designed to compare the similarity between two inputs by learning a shared embedding space, offer robust capabilities in identifying related entities based on limited or noisy data. Our approach enhances security by combining the unique properties of PUFs with the comparison abilities of SNNs, reducing computational overhead and storage requirements. The results show that the proposed method outperforms traditional authentication techniques in accuracy and resilience, offering a scalable solution for secure user and device authentication.