Text Detection of Power Nameplate Based on Improved DBNet

Aiming at the problem of low efficiency and low accuracy of power nameplate text detection in natural scenarios, we propose a detection model based on improved DBNet. By introducing the MobileOne network, the feature extraction ability of the model is enhanced, while ensuring the real-time nature of the model, and the Feature Pyramid Network is improved, and the RepResBlock is introduced to control the amount of information in the previous layer, and at the same time, combined with the attention mechanism, improve the model's attention to text features, and strengthen the model's ability to integrate and express multi-scale features. Experimental results show that the model can maintain good detection results under small parameters. Compared with the original DBNet model, our model has significantly improved performance, achieving more accurate, more robust and faster power nameplate detection.