In Morocco, the challenges of road safety and traffic management are exacerbated by the growing number of vehicles. Automatic detection and recognition of number plates is essential for many applications such as traffic management, access control and surveillance systems. Advances in deep neural networks, particularly You Only Look Once (YOLO) object detection models, have been widely used for this task. Recently, Ultralytics presented an improved version of YOLO with better performance. This research proposes a new approach based on YOLOv8, integrating attention mechanisms into the original architecture. We have integrated and evaluated four attention modules: CBAM, GAM, ECA and SA within YOLOv8 to design improved models for the detection of Moroccan patches. These models, were trained and evaluated on a set of 1600 images captured under various conditions. The results show that Our model based on ResBlock+CBAM (ResCBAM) achieves top performance with an average mAP@0.5 of 99,7%, setting a new record. In addition, the Tesseract OCR engine was used to recognise the characters on the plates detected. Its robustness and text extraction accuracy make it an excellent complement to the proposed models. This integration provides a complete license plate detection and recognition system. The study confirms the effectiveness of attention mechanisms and OCR technology such as Tesseract in improving performance, which is crucial to traffic management and road safety in Morocco.