A hyperspectral image classification method based on meta-pseudo-label multi-scale graph convolutional network

By adopting a multi-scale graph convolution network method based on metapseudo-label in hyperspectral image classification, the problem of performance bottleneck in complex geographic classification tasks is solved, and higher classification accuracy and robustness are achieved, especially in small samples and scarce labels.

CN119399562BActive Publication Date: 2025-03-25ZHEJIANG COLLEGE OF SECURITY TECH +2
2 Cites 0 Cited by

Patent Information

Application Number
CN202510004147.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2025-01-02
Publication Date
2025-03-25
Estimated Expiration
2045-01-02

AI Technical Summary

Technical Problem

Traditional hyperspectral image classification methods have performance bottlenecks when dealing with complex geographic classification tasks, and cannot fully tap into potential information in hyperspectral data, and it is difficult to effectively capture key features when labels are scarce.

Method used

A multi-scale graph convolutional network (MPL-MFSGCN) method based on meta-pseudo-label is used to extract pixel-level features through convolutional neural networks, perform superpixel segmentation, and build a multi-scale feature search graph convolutional network, combining meta-pseudo-labels and teacher-student models for semi-supervised learning, and optimize network parameters to improve classification accuracy.

Benefits of technology

It significantly improves the ability to extract irregular land features, improves the classification performance in small samples and scarce labels, and can make full use of unlabeled data for training, improving classification accuracy and robustness.

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

Abstract

The present invention relates to a hyperspectral image classification method based on a meta-pseudo-label multi-scale graph convolutional network. By establishing a meta-pseudo-label multi-scale feature search graph convolutional network MPL-MFSGCN, the HSI is first divided into super-pixel graphs of different scales and converted into a graph structure using the SLIC super-pixel segmentation technology; then a multi-scale feature search graph convolutional network MFSGCN architecture is constructed, and a neural network search algorithm is used to optimize the super-pixel feature weight combination to enhance the discriminative power of feature expression; meta-pseudo-labels are generated by adopting a semi-supervised teacher-student model that shares the same graph convolutional network, and the student network is fine-tuned to improve the robustness of the network; key features can be automatically captured under small sample conditions, and higher classification accuracy than advanced methods is achieved on commonly used HSI datasets, effectively improving the classification accuracy.
Need to check novelty before this filing date? Find Prior Art