An optical remote sensing ground feature relationship semantic understanding system and method in a localization environment

Abstract

The application discloses an optical remote sensing ground object relationship semantic understanding system and method in a domestic environment, constructs a remote sensing scene relationship analysis module, a multi-modal scene knowledge base module, a remote sensing scene graph construction module, a remote sensing scene representation module and a multi-modal sample pair construction module, uses a code large language model suitable for a domestic platform to programmatically model ground object relationships, and generates scene graph triples; through a double-branch double-time phase comparison learning network, time sequence vision and structural knowledge are fused to realize cross-modal joint representation learning. The application solves the problems of incomplete ground object relationship expression, inaccurate modeling and lack of real-time semantics in the prior art, and deeply optimizes data labeling, heterogeneous computing power management and the like in the domestic environment, and significantly improves the accuracy, completeness of ground object relationship semantic understanding and the running efficiency on the domestic software and hardware platform.

Images