A building extraction method for high-resolution remote sensing images

CN122156669APending Publication Date: 2026-06-05CHINA THREE GORGES UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
CHINA THREE GORGES UNIV
Filing Date
2026-02-09
Publication Date
2026-06-05

AI Technical Summary

Technical Problem

Existing deep learning-based building extraction methods face challenges when processing high-resolution remote sensing images. These challenges include incomplete extraction due to the complex structure and varied shape and scale of buildings, the omission of small buildings, and the loss of image detail information caused by continuous downsampling in deep learning, which affects the accuracy of boundary extraction.

Method used

We construct a lightweight building extraction network, LBEHRNet, which employs a lightweight star-shaped convolutional module, a cross-resolution assisted hybrid attention module, and a multi-resolution fusion prediction head. Through multi-stage feature extraction and fusion, it enhances the ability to extract building boundary details and multi-scale features, while reducing the number of model parameters and computational complexity.

Benefits of technology

While reducing model complexity, the accuracy and efficiency of building extraction were improved, enabling efficient and accurate extraction and visualization of building areas, and verifying the effectiveness and practicality of the model in actual high-resolution remote sensing image scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122156669A_ABST
    Figure CN122156669A_ABST
Patent Text Reader

Abstract

A building extraction method for high-resolution remote sensing images comprises the following steps: step 1, collecting a building extraction data set of high-resolution remote sensing images, performing image preprocessing and data enhancement; step 2, performing building region pixel-level labeling on remote sensing images in the data set, and dividing a training set, a verification set and a test set according to a preset ratio; step 3, constructing a light building extraction network LBEHRNet for generating a building segmentation result graph; step 4, inputting the data set into the LBEHRNet network for training, combining a segmentation model to perform building extraction on high-resolution remote sensing images, and using precision Pre, recall Rec, F1 value, intersection-over-union IoU, parameter quantity Params and floating-point operation number FLOPs as evaluation indexes; and step 5, using the trained model to perform building extraction on remote sensing images to be predicted, and outputting a building segmentation result graph. The method can be well used for high-precision extraction of remote sensing image buildings.
Need to check novelty before this filing date? Find Prior Art