Topographic map unmarked region segmentation method based on deep learning

CN120807928AActive Publication Date: 2025-10-17CHINA AERO GEOPHYSICAL SURVEY & REMOTE SENSING CENT FOR LAND & RESOURCES
2 Cites 0 Cited by

Patent Information

Application Number
CN202510934498.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-07-08
Publication Date
2025-10-17
Estimated Expiration
2045-07-08

AI Technical Summary

Technical Problem

Existing methods for segmenting unlabeled regions in topographic maps are inefficient, inaccurate, have poor generalization capabilities, lack spatial coherence and terrain structure constraints, and are difficult to determine the correctness of the segmentation results.

Method used

A deep learning-based unlabeled region segmentation method for topographic maps is adopted. By integrating topographic maps, DEM, remote sensing images and OSM vector data as multi-source tensor data, a backbone encoder and segmentation network are designed to perform self-supervised tasks and position encoding. A geographic semantic short sentence database is introduced, prediction uncertainty is quantified, and multiple rounds of iterative training are performed. Pixel connected domain detection and terrain constraints are combined to generate fine segmentation results.

Benefits of technology

It achieves precise segmentation of complex landforms, improves segmentation efficiency and accuracy, enhances the generalization ability and spatial consistency of the model, significantly optimizes the boundaries and spatial distribution of unlabeled areas, and promotes the intelligent upgrade of geographic information systems.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120807928A_ABST
    Figure CN120807928A_ABST
Patent Text Reader

Abstract

The invention provides a topographic map unmarked region segmentation method based on deep learning, and the method comprises the steps: collecting and segmenting a topographic map grid, and obtaining a multi-source tensor topographic map; designing a backbone network, and outputting feature representation of the multi-source tensor topographic map; constructing a geographic semantic library, increasing geographic semantic understanding of the topographic map, and generating a semantic weight; designing a trunk segmentation network, and performing initial segmentation prediction; quantizing the prediction uncertainty of the trunk segmentation network, and outputting an initial segmentation result through a multi-round iteration mode; pixel connected domain detection, terrain constraint and space consistency improvement are carried out on the initial segmentation result, and a coherent segmentation result is obtained; and then a coherent segmentation result is visualized into a heat map and a spatial interpretability map layer. According to the method, the complex landform and spatial isomerism conditions can be processed, the segmentation of the unmarked region of the topographic map is more accurate, intelligent and efficient, the real landform of the unmarked region is restored, the boundary and spatial distribution is remarkably optimized, and the intelligent upgrading of a geographic information system is promoted.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Map automatic updating method based on remote sensing image, storage medium and system

    CN113204608A

  • Boundary-optimized remote sensing image semantic segmentation method and apparatus, and device and medium

    WO2023077816A1