Adversarial attacks are designed to trick deep neural networks, leading to incorrect predictions of results and potential security risks. Although extensive research has been conducted on adversarial attacks in various applications in the computer vision domain, there is insufficient research on the impact of such attacks on license plate recognition (LPR) systems. The purpose of this work is to fill this gap by using the FGSM attack as a first-generation attack and the DI2-FGSM attack as a modern attack on our YOLOv5 and CnOCR-based LPR system. In addition, we are exploring PGD-based adversarial training as a mitigation method for the LPR system. The objectives of this research include studying various types of adversarial attacks, evaluating their applicability to LPR systems, conducting attacks based on a specific model, comparing attack success rates and evaluating the effectiveness of adversarial learning as a mitigation method. This research has implications for improving the safety and reliability of LPR systems in critical applications such as traffic regulation, law enforcement and accident investigation. The current paper sets the ground for extensive research.