Multi-scale Vehicle Logo Recognition based on Self-attention Mechanism and YOLOv8

The vehicle logo, as one of the key features for vehicle identification, is of great significance in the aspect of assisting the management of intelligent transportation system. It has significant application value and practicality in scenarios such as traffic management, vehicle tracking, intelligent driving assistance, and violation monitoring. Due to the condition of the logo itself, photographing circumstance restrictions, and light intensity, the vehicle logo is not easily recognizable. Vehicle logo recognition remains a research hot topics in the field of computer vision. This paper explores the design and implementation improved YOLOv8 for vehicle logo recognition. The method combines self-attention mechanism with YOLOv8, improving the efficiency and accuracy of vehicle logo recognition in complex traffic environments. The modified YOLOv8 can quickly and accurately identify vehicle logos in various scenarios. The experiment results showed that the average accuracy (mAP@o.5) of the model on the training set reached 97.28%, and the precision and recall rates reached 95.52% and 95.16% respectively. On the validation set, the recognition accuracy reached 95.6%, the recall rate reached 95.0%, and the average accuracy (mAP 50) reached 96.4%. The algorithm model has a good detection effect on vehicle logo target recognition.