Automatic Train Number Recognition Based on Dual Attention Mechanism and Multi-scale Feature Fusion

This study is committed to improving the YOLOv5 model to improve the positioning and recognition accuracy of train car numbers in natural environments. In today’s railway transportation system, automatic identification of car number plays a vital role in improving logistics efficiency and safety. In the face of complex background and changeable interference factors, traditional recognition methods often perform poorly, and the introduction of deep learning technology provides a new perspective to solve this problem. In this study, by adding the Dual Attention mechanism (DANet) and multi-scale feature fusion strategy to the YOLOv5 model, the adaptability of the model in diverse natural environments and the recognition accuracy of objects of different sizes are significantly improved. In order to verify the effectiveness of the improved strategy, we collect and conduct experiments on a self-made dataset of 4000 high-resolution train car number images, which cover a wide range of natural scenes. Experimental results show that the improved YOLOv5 model exceeds a variety of advanced object detection models in key performance indicators such as precision, recall rate and F1 score, reaching a recognition accuracy of up to 99%. In addition, by performing ablation experiments, the key role of DANet and cross-scale feature fusion in improving the performance of the model is further confirmed. The results of this research not only greatly promote the development of automatic recognition technology of railway car numbers, but also provide valuable experience and methods for the application of deep learning technology in traffic monitoring systems and other related fields.