多模态场景分析及地理位置预测技术研究

Considering multi-modality of scene data, multi-level nature of semantics and multi-granularity of geographical location, towards three major scientific issues on scene-oriented multi-modal data representation and fusion, multi-level semantic analysis and understanding, multi granularity location inference and prediction, this project works on the key techniques of scene-oriented visual representation learning, multi-modal embedding and learning, multi-level semantic joint learning, geographical location inference with knowledge graph, and multi-granularity geographical location prediction. This project will create a set of theories and techniques for multi-modal scene data in representation, association, understanding, inference and prediction, and will also develop a technical framework, ranging from analyzing and understanding multi-modal scene data to inferring and predicting multi-granularity geographic location. Based on the technical framework, we plan to develop the applications including multimedia data geographical location annotation, organization, retrieval and visualization to verify the effectiveness of the proposed theory, methods and technologies.

本课题针对场景数据的多模态性、场景语义的多层次性和场景位置的多粒度性等特点，围绕面向场景的多模态数据表示和融合、多层次语义分析和理解、多粒度位置推断和预测等三个关键科学问题，研究面向场景的视觉表示学习、多模态嵌入与学习、多层次语义联合学习、融入知识图谱的地理位置推断和多粒度地理位置预测等技术，形成一套多模态场景数据的表示、关联、理解、推断和预测的理论体系，实现从多模态场景分析和理解到多粒度地理位置推断和预测的技术框架。在此基础上开展多媒体数据的地理位置标注、组织、检索及可视化等技术验证，验证所提理论、方法与技术的有效性。

内容获取失败，请点击重试