Lightweight deep learning-driven secure communication architecture for Internet of Things enabled government applications

Internet of Things (IoT)-based devices are extensively utilized for data transmission to the cloud across various organizations. Nonetheless, there are notable limitations in the conventional approach, such as in critical situations, where transmitting sensitive data, secure communication across the cloud is not guaranteed, and the memory, processing power, and bandwidth constraints in these IoT devices present significant challenges. The suggested model employs a bespoke Convolutional Neural Network (CNN) to categorize sensitive and non-sensitive images on the device, utilizes Elliptic Curve Cryptography (ECC) for safe session key sharing, implements SHA-512 hashing for integrity verification, and applies the ChaCha20 stream cipher for rapid, random encryption for sensitive images. The mean entropy of the proposed technique is 7.9976, and the correlation coefficients approximate zero. The histogram distributions are balanced, rendering statistical attacks exceedingly difficult. This approach surpasses AES + RSA-1024, SPECK, and PRESENT by reducing the average encryption time by up to 99%, enhancing throughput by over 647%, and consuming up to 99.79% less energy. This proposed solution offers a robust, efficient, and secure framework for managing sensitive government data, effectively addressing both the resource constraints of IoT devices and the necessity for privacy in governmental communication systems.