A hyperspectral remote sensing image classification method and system based on a fusion graph network

By using a fusion graph network-based approach, combining superpixel and pixel features to construct a graph structure and perform feature interaction fusion, the problem of global and local feature extraction in hyperspectral remote sensing image classification is solved, achieving higher classification accuracy.

CN118781403BActive Publication Date: 2026-06-19WUHAN UNIV
2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
WUHAN UNIV
Filing Date
2024-06-26
Publication Date
2026-06-19

AI Technical Summary

⚠Technical Problem

Existing hyperspectral remote sensing image classification methods struggle to effectively combine pixel spectral features and spatial relationships when processing large-scale remote sensing images, resulting in limited classification accuracy. Furthermore, existing algorithms fail to effectively extract global and local features, impacting classification performance.

⚗Method used

A method based on fusion graph networks is adopted. The superpixels are divided and graph structures are constructed by using the HSLIC algorithm. By combining spectral and spatial features, multiple pixel subgraph structures are constructed. Single-layer graph convolutional network feature transformation and feature interaction fusion are performed to optimize graph structure and node features, and achieve mutual guidance of global and local features.

🎯Benefits of technology

It improves the classification accuracy of hyperspectral remote sensing images, enabling better capture of global and local features in the images, thus enhancing the accuracy and efficiency of classification.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

This invention provides a hyperspectral remote sensing image classification method and system based on a fusion graph network, comprising: constructing initial graph structures for single-pixel branches and superpixel branches according to the spectral features of the dataset; inputting all graph structures from the two branches into a single-layer graph convolutional network to obtain output features; concatenating the output features of multiple sub-graphs in the single-pixel branch to restore the original graph structure features; reassigning the superpixel features in the superpixel branch to restore the original graph structure features; inputting the restored features into an interactive fusion module to fuse the features extracted from the two branches; redistributing the updated features into sub-graphs and superpixel graphs; concatenating the features from the two branches and passing them through an output layer to obtain the final classification result. This invention improves the superpixel segmentation algorithm to take into account the characteristics of hyperspectral remote sensing images, enabling it to consider global information in the image while also taking into account feature extraction of both coarse and fine granular information.
Need to check novelty before this filing date? Find Prior Art