Most existing license plate (LP) detection and recognition studies are evaluated on single-region datasets, despite the availability of multiple large, diverse datasets from various countries. In this project, we implement YOLO models to detect license plates using datasets from different regions, as well as a combined dataset that encompasses a wide variety of LP patterns. The datasets include images captured under diverse conditions, ranging from manual collection by roadside parking management personnel to automatic capture by roadside cameras. By comparing models trained on individual datasets and the combined dataset, we analyze how the diversity of LP patterns impacts model accuracy and generalization.