Scene text recognition method for guiding attention generation based on cross-domain supervision signal

The invention discloses a scene text recognition method for guiding attention generation based on a cross-domain supervision signal. The method comprises the following steps of: extracting a text core region as a supervision signal in a coding stage, recursively performing attention guidance, fusing coding information generated by guiding attention with coding information generated without guiding attention by using a gating mechanism, and enhancing the robustness of the coding information; in the decoding stage, an efficient and parallel adaptive conversion decoder is combined for decoding, attention offset in the decoding stage is prevented, and the recognition performance of the model is improved. Besides, in a training stage, the method adopts a fusion strategy of artificial guidance and model adaptive learning to accurately learn a core region of a text, so that a correct supervision signal is provided for attention guidance. 本发明公开了基于跨域监督信号引导注意力生成的场景文本识别方法。该方法在编码阶段提取文本核心区域作为监督信号递归地进行注意力引导，使用门控机制将引导注意力生成的