Design and Implementation of a Deep Learning Embedded Cryptosystem for IoT Applications

Securing communications in the Internet of Things (IoT) is a critical challenge, particularly for resource-constrained devices operating on local networks. This paper presents a comprehensive cryptosystem for text-based message transmission between an IoT device and a connected server. We propose a deep learning embedded cryptosystem that integrates several advanced security techniques. Our system leverages the DiffieHellman (DH) key exchange to securely establish a shared secret between the server and devices, which is then used with Advanced Encryption Standard in Galois/Counter Mode (AESGCM) encryption scheme to safeguard messages. JSON Web Token (JWT)-based authentication ensures that only devices with valid credentials can access the system. At the same time, a Convolutional Neural Network (CNN)-based machine learning model further analyzes device-specific logs to classify devices as trusted or malicious. Qt-based applications on both the IoT device and the server facilitate real-time interaction, monitoring, and control. Experimental results demonstrate key exchange times of 150 to 200 ms, encryption/decryption latencies of approximately 50 ms per message, and JWT verification within 20 ms, while the CNN model achieves over 90 % classification accuracy, with precision, recall, and F 1 -scores of approximately <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$95.2 %, 94.7 %$</tex>, and 94.9 %, respectively. This integrated solution not only provides robust security but also lays the groundwork for future real-time anomaly detection and scalable IoT deployments.