Remote sensing image water semantic segmentation method based on feature extraction and edge details

By introducing dual attention gating and deformable convolutional decoding blocks into water body segmentation of remote sensing images, combined with a boundary refinement module, the problems of noise interference and boundary information loss in water body segmentation of remote sensing images are solved, achieving higher segmentation accuracy and robustness.

CN122391869APending Publication Date: 2026-07-14LANZHOU UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
LANZHOU UNIV
Filing Date
2026-04-20
Publication Date
2026-07-14

AI Technical Summary

Technical Problem

Existing water segmentation methods suffer from insufficient selectivity of skip connection features, poor adaptability of fixed receptive fields, and loss of boundary information in remote sensing images, resulting in insufficient segmentation accuracy and robustness.

Method used

We employ a dual attention gating mechanism to filter noise, use deformable convolutional decoding blocks to adaptively extract water features, enhance boundary accuracy through a boundary refinement module, and improve model stability and generalization performance by combining K-fold cross-validation and test-time enhancement strategies.

Benefits of technology

It significantly improves the boundary accuracy and visual quality of water body segmentation, solves the problems of noise interference and loss of boundary information, and improves the segmentation accuracy and robustness of the model.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122391869A_ABST
    Figure CN122391869A_ABST
Patent Text Reader

Abstract

The application relates to a remote sensing image water body semantic segmentation method based on feature extraction and edge details, which comprises the following steps: acquiring remote sensing image data and corresponding label data, and sequentially performing attribute reading, data verification and screening and data division processing on a data set composed of the remote sensing image data and the corresponding label data to obtain a local database; forming an initial water body semantic segmentation model for remote sensing image water body semantic segmentation; constructing a training mode library; performing water body segmentation training, verification and test processing on the initial water body semantic segmentation model to obtain a final water body semantic segmentation model; and performing enhanced inference prediction on the target image and the images after horizontal flipping, vertical flipping and bidirectional flipping of the target image to obtain a water body semantic segmentation image. The application can remove noise in shallow features in advance, and only pure target details are sent into a decoder.
Need to check novelty before this filing date? Find Prior Art