Method and system for detecting camouflaged targets based on hierarchical graph interaction and unit clustering

Through the camouflage object detection method of hierarchical graph interaction and unit clustering, the clustering algorithm is used to remove redundant feature map units and perform self-attention mechanism calculations, the problems of poor performance and unclear details of the existing model in complex scenarios are solved, and a higher precision camouflage object detection is achieved.

CN119206168BActive Publication Date: 2025-07-08BEIJING UNIV OF POSTS & TELECOMM
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202411219639.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-02
Publication Date
2025-07-08
Estimated Expiration
2044-09-02

Smart Images

  • Figure CN119206168B_ABST
    Figure CN119206168B_ABST
Patent Text Reader

Abstract

The present invention provides a camouflaged target detection method and system based on hierarchical graph interaction and unit clustering. The method includes: obtaining an image to be detected, converting it into a set number of feature map units through unit division and linear transformation, inputting all the feature map units into multiple consecutive region-aware unit attention-focusing modules, removing redundancy based on the similarity of the self-attention mechanism using a clustering algorithm, performing self-attention mechanism calculation on the filtered feature map units, and outputting a first-level feature map. Pairwise input the first-level feature maps output by adjacent region-aware unit attention-focusing modules into a hierarchical graph interaction self-attention module, convert them into a graph structure for hierarchical feature interaction, and project the interacted graph structure back to the original space to output a second-level feature map. Input the second-level feature map into a confidence aggregation feature fusion decoder to refine ambiguous regions and fuse all the second-level feature maps, and output a camouflaged target. The present invention can improve the detection accuracy.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Sketch semantic segmentation method based on class coding and decoding structure

    CN114998904A

  • Hyperspectral image subspace clustering method and system

    CN118351341A