Rozpoznání značek nebezpečí v bezpečnostní praxi s pomocí AI

Abstract

This thesis deals with the application of deep learning methods for the recognition of hazard signs in the field of security practice. The theoretical part focuses on physical security, safety signage, and the fundamentals of artificial intelligence, with an emphasis on convolutional neural networks. In the practical part, selected neural network architectures (GoogLeNet, ResNet18, and ResNet50) are implemented and evaluated for image classification. A series of experiments with different hyperparameter settings was carried out to analyze their impact on classification performance. The results show that the best performance is achieved by the ResNet50 model, reaching a maximum validation accuracy of 99.59 %. However, this model is also the most computationally demanding. The ResNet18 model represents a suitable compromise between classification accuracy and computational efficiency.