Abstract: Aiming at the problem that the traditional recurrent neural network (RNN) structure in  image Chinese caption was not conducive to long sentence generation and lacked detailed semantic information, we proposed a Transformer multi-head attention (MHA) network, which  fused the coarse-grained global features and fine-grained regional target entity features.  Through the fusion of multi-scale features,  the method made it easier for image attention to focus on fine-grained target regions and an image representation with more fine-grained semantic features was obtained, thus effectively improving image caption. A variety of evaluation indicators were used for verification on the ICC dataset, the results show that the model achieves better image caption effects in all indicators.

Key words: image Chinese caption, fine-grained feature, multi-head attention (MHA)