Využití generativní umělé inteligence pro vytváření syntetických obrazů zbraní a střeliva

Abstract

This thesis explores the use of generative artificial intelligence to create synthetic images of weapons and ammunition, with a focus on their application in the field of physical security engineering. The aim of this work is to design and implement a generative adversarial network (GAN) model capable of generating realistic visual representations of selected types of weapons and subsequently evaluate the quality of these outputs. As part of the thesis, a custom image dataset was created and used to train a DCGAN (Deep Convolutional Generative Adversarial Network) model in the MATLAB environment. The quality of the generated images was evaluated using non-reference metrics such as NIQE, BRISQUE, PIQE, entropy, sharpness, and contrast. The contribution of this work is the verification of the usability of generative models for creating synthetic data, which can serve as a supplement to real-world datasets in the design and testing of security systems focused on the detection of dangerous objects.