Image-text retrieval method and system based on attention mechanism and gating mechanism

The invention discloses an image-text retrieval method and system based on an attention mechanism and a gating mechanism, and belongs to the field of cross-modal retrieval. According to the invention, valuable information is rapidly screened out through an attention mechanism to obtain more accurate feature expression, on the basis, in order to make the corresponding relation between the modals more obvious, data of the two modals serve as supervision information mutually, a gating mechanism is introduced to further adjust the feature of the other modal, so that unnecessary information is filtered as much as possible, and parts with rich semantics are reserved; and finally, image features with enough general semantics and accurate attention are obtained, so that the performance of a cross-modal retrieval model is effectively improved. 本发明公开了一种基于注意力机制和门控机制的图文检索方法和系统，属于跨模态检索领域。本发明通过注意力机制快速筛选出有价值的信息，以获取更加精确的特征表达，在此基础上，为了使模态之间的对应关系更加显著，通过将两种模态的数据互相作为监督信息，并引入门控机制对另一模态特征进行进一步的调整，以尽可能多地过滤掉不必要的信息，保留语义丰富的部分，最终获得具有足够通用