Building extraction method from remote sensing image based on fractal geometric features and edge supervision

By combining fractal geometric features and edge supervision with the FB-Unet network model, the problem of incomplete edge extraction of small-scale and irregular buildings in remote sensing images was solved, and higher-precision building edge segmentation was achieved.

CN116543298BActive Publication Date: 2025-10-24SOUTHEAST UNIV
View PDF -1 Cites 0 Cited by

Patent Information

Application Number
CN202310347368.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-03-30
Publication Date
2025-10-24
Estimated Expiration
2043-03-30

AI Technical Summary

Technical Problem

Existing building segmentation methods for remote sensing images have difficulty in effectively extracting the edges of small-scale buildings and irregular buildings, resulting in incomplete edges and unable to ensure the integrity of building edges and overall structural similarity.

Method used

The FB-Unet network model based on fractal geometric features and edge supervision is adopted. The network is trained with training set samples and the segmentation accuracy of building boundaries is enhanced by using multi-scale expansion-fractal geometric feature module and edge supervision network.

Benefits of technology

It improves the segmentation accuracy of building edges, enhances the extraction effect of buildings in complex backgrounds, and ensures the integrity and structural similarity of building edges.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116543298B_ABST
    Figure CN116543298B_ABST
Patent Text Reader

Abstract

The application discloses a kind of based on fractal geometric features and edge supervision's remote sensing image building extraction method, comprising the following steps: obtaining a large amount of remote sensing image data constructs image data set, obtains the building binary graph label corresponding to each remote sensing image, that is real label;Obtain a large number of sample data, and all samples are proportionally divided into training set, verification set and test set;Image in training set sample is used as the input of FB-Unet network, label in training set sample is used as the true value label of FB-Unet network, FB-Unet network is trained, and building extraction network FB-Unet model is obtained after training is completed;The remote sensing image to be carried out building extraction is input into the building extraction network FB-Unet model trained, extracts the semantic feature of building in remote sensing image, and obtains the pixel-by-pixel prediction result corresponding to remote sensing image.The application solves the problem that building edge reservation is not complete, and the extraction effect of irregular building is poor.
Need to check novelty before this filing date? Find Prior Art