A remote sensing change detection method and system based on a multi-modal base model and a text condition bridge

CN122265855APending Publication Date: 2026-06-23JIANGSU WATER CONSERVANCY SCI RES INST +1
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-05-28
Publication Date
2026-06-23

AI Technical Summary

Technical Problem

Existing remote sensing change detection methods struggle to distinguish between target changes and false changes in complex scenarios, and lack the ability to dynamically adjust to user semantic intent, resulting in high false alarm rates and insufficient generalization.

Method used

A method based on multimodal base model and text conditional bridging is adopted. Features are extracted by pre-trained visual base model and text encoder, combined with text semantic embedding vector for dynamic semantic modulation and directional perception difference modeling, generating target change difference features, and outputting change detection results in decoder.

Benefits of technology

It significantly improves the flexibility and accuracy of remote sensing change detection, reduces training costs, enhances the robustness and interpretability of the model, and can accurately identify specified change categories, attributes and locations, thereby reducing the false alarm rate.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122265855A_ABST
    Figure CN122265855A_ABST
Patent Text Reader

Abstract

The application discloses a remote sensing change detection method and system based on a multi-modal base model and a text condition bridge, and belongs to the technical field of remote sensing image processing. The method comprises the following steps: acquiring double-time-phase remote sensing images of the same geographical area at different times and corresponding natural language text instructions; extracting a double-time-phase visual feature sequence by using a pre-trained visual base model, and extracting a global text semantic embedding vector by using a text encoder; constructing a text condition bridge module, performing dynamic semantic modulation on visual features by using text semantics, and generating adaptive features consistent with text semantics; constructing a directional perception difference modeling module, combining text semantics to generate target change difference features; and finally outputting a change detection result corresponding to the text instruction through a change detection decoder. The application introduces text semantic constraints in the knowledge transfer stage of the base model, reduces the injection of general knowledge irrelevant to the target change, suppresses the pseudo-change response, and realizes high-precision detection of specified changes.
Need to check novelty before this filing date? Find Prior Art