A Collaborative Extraction Algorithm for Multi-Source Remote Sensing Land Parcel Boundaries Based on FedGNN

View at publisher Summarise with VIRAT
Authors: Xingyao Liu, Qingxiang Meng
Cited by: 0
References: 8

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

In multi-source remote sensing images of land parcels, only the background feature map can be extracted, resulting in low accuracy of boundary extraction. Therefore, a collaborative extraction algorithm for multi-source remote sensing land parcel boundaries based on FedGNN is proposed. In the preprocessing stage of the multi-source remote sensing images, median filtering is first used to reduce noise and preserve edge information. Then, a Gaussian mixture model is used to optimize pixel distribution, and adaptive grayscale transformation is used to enhance contrast. An adapted FedGNN training strategy is constructed, using dynamic weighted sample selection. Local model outputs are aggregated to obtain a global feature map, and a fusion function is defined to fuse land parcel background and boundary information. A unified feature scale is applied to the multi-source remote sensing images to reduce computational complexity and improve local feature extraction accuracy. B-spline curve fitting based on curvature constraints is used to optimize boundary geometric accuracy. A dynamic threshold is defined to distinguish between boundary and non-boundary regions, ultimately obtaining a multi-source remote sensing land parcel boundary map with topological consistency, achieving collaborative extraction of multi-source remote sensing land parcel boundaries. Experimental results show that the designed algorithm is accurate and stable in extracting the boundaries of regular maize planting areas, and also has certain effectiveness in complex soybean terrace areas, providing reliable technical support for collaborative extraction. In terms of accuracy evaluation, after processing 1200 images, the accuracy ($\mathbf{P}$) reached 0.90, the recall (R) increased to 0.86, the false detection rate (FDR) decreased to 0.09, and the false negative rate (FNR) decreased to 0.14. In the boundary continuity comparison, the designed algorithm scored as high as 90, far exceeding the$50-64$of references [1]–[4], which can more completely and accurately delineate the outline of the plot, and the boundary extraction accuracy is high.

Fields of Study & Subjects

Remote-Sensing Image Classification Automated Road and Building Extraction Remote Sensing in Agriculture

Publication details

Journal / source
International Conference on Smart City and Information System (ICSCIS)
Pages
19–22
Publication date
30 January 2026
Language
EN