However, these methods lack guidance and constraints during the extraction and fusion of multimodal features and do not fully leverage the complementary advantages between global and local features, ...