DEEP LEARNING FOR SECURITY SYSTEMS FROM CAMOUFLAGED THREAT DETECTION TO QUANTUM KEY PROTECTION

Abstract

Because today’s world is so interconnected and digital, it is now even more crucial to keep both physical and digital systems secure. With rapid growth and practical uses, deep learning is becoming very important in facing current security difficulties. Aspects of system security such as camouflaged object detection and quantum communication protection, are discussed in this thesis. Detecting camouflaged objects remains a difficult problem. Many domains rely heavily on the usefulness of this field. It happens because the object we want to find looks similar to its background. There are many strategies and datasets being developed to deal with this issue and this field has emerged as a rapid growth point in image processing. We tested EfficientDet with SAM on NC4K and compared the results to what some existing models show. By analysing why the model failed, we have suggested areas for improvement in future projects. In this part, the thesis compares EfficientDet and SAM to various COD models and also examines how the new NC4K dataset performs. This thesis further examines the topic of system security by exploring how deep learn ing can support quantum communication. In the second section, we study the use of neural networks for error correction in QKD. To study five architectures, a new dataset of 120,000 observations was created, where both noise probabilities and photon transmis sion rates varied. From these findings, it is clear that AI can improve exiting quantum communication protocols.