A community natural disaster damage degree assessment method and related device

By explicitly modeling the relationships between local areas within a community, and using a backbone feature extraction network and learnable region query vectors for adaptive region aggregation, the problem of local damage information being difficult to transmit to the whole community in community-level scenarios is solved, achieving higher assessment accuracy and stability.

CN122435489APending Publication Date: 2026-07-21HAINAN UNIV
0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
HAINAN UNIV
Filing Date
2026-04-28
Publication Date
2026-07-21

Smart Images

  • Figure CN122435489A_ABST
    Figure CN122435489A_ABST
Patent Text Reader

Abstract

The application discloses a community natural disaster damage degree evaluation method and related device. The method comprises the following steps: obtaining a community disaster image; extracting an image feature map in the community disaster image through a backbone feature extraction network, performing tiling processing on the extracted image feature map, and obtaining a feature sequence; performing adaptive region aggregation on the feature sequence by using a learnable region query vector, generating a plurality of local region features and corresponding region center positions; constructing inter-region semantic relationship and spatial relationship based on the local region features and the region center positions, and fusing to form a comprehensive region relationship matrix, performing relationship propagation update on the local region features, and obtaining the local region features after relationship enhancement; and performing importance weighted aggregation on the local region features after relationship enhancement, and generating a community overall damage representation. The application can improve the accuracy and stability of community disaster damage evaluation.
Need to check novelty before this filing date? Find Prior Art