Symmetric Keys for Lightweight Encryption Algorithms Using a Pre–Trained VGG16 Model

Abstract

The main challenge within lightweight cryptographic symmetric key systems is striking a delicate balance between security and efficiency. Consequently, the key issue revolves around crafting symmetric key schemes that are both lightweight and robust enough to safeguard resource-constrained environments. This paper presents a new method of making long symmetric keys for lightweight algorithms. A pre–trained convolutional neural network (CNN) model called visual geometry group 16 (VGG16) is used to take features from two images, turn them into binary strings, make the two strings equal by cutting them down to the length of the shorter string, and then use XOR to make a symmetric key from the binary strings from the two images. The key length depends on the number of features in the two images. Compared to other lightweight algorithms, we found that this method greatly decreases the time required to generate a symmetric key and improves defense against brute force attacks by creating exceptionally long keys. The method successfully passed all 15 tests when evaluated using the NIST SP 800-22 statistical test suite and all Basic Five Statistical Tests. To the best of our knowledge, this is the first research to explore the generation of a symmetric encryption key using a pre–trained VGG16 model.

Keywords

• lightweight cryptographic
• transfer learning
• pre–trained VGG16
• feature extraction