In response to the challenges of information loss and omission in text detection within complex natural scenes,the Attention mechanism is introduced and the Attention-DBNet algorithmis proposed.The Attention mechanism is incorporated into the FPNNet(Feature Pyramid Networks)architecture to enhance the feature extraction capabilities of the backbone network,enabling the model to focus on relevant information while suppressing irrelevant details.Additionally,a novel binarized differentiation formula is introduced during the model's prediction stage to achieve more accurate pixel-level classification and expedite model training convergence.Experimental results demonstrate the superiority of the Attention-DBNet algorithm over other state-of-the-art methods on multiple datasets,with recall,precision,F-measure,and detection time all exhibiting an improvement of over 10%.
text detectionattention mechanismDBNet algorithmdifferential formula
维普
万方数据
