Image-text retrieval method and system based on feature enhancement

The invention provides an image-text retrieval method and system based on feature enhancement, and relates to the field of cross-modal retrieval, the specific scheme comprises: obtaining a to-be-retrieved sample, the to-be-retrieved sample being a to-be-retrieved image or a to-be-retrieved text; inputting a to-be-retrieved sample into the trained cross-modal feature extraction model to obtain retrieval features mapped to the same public semantic space; the similarity degree between the retrieval features is judged through the distance, and then a retrieval result is obtained; text branches and image branches are adopted to extract images and text features, the two branches learn feature relationships between paired text samples and image samples, Transformer encoders of the image branches increase feature enhancement attention, image features are enhanced from two different dimensions of channels and spaces, redundant information and noise in the image features are filtered, and the image features are extract