Remote sensing image semi-supervised semantic segmentation method, device, equipment and medium

By constructing a semi-supervised semantic segmentation model for remote sensing images, and utilizing feature extraction from labeled and unlabeled data and prototype loss constraints, the problem of class imbalance in remote sensing images is solved, and the model's ability to identify and segment a minority of classes is improved.

CN121982576BActive Publication Date: 2026-06-23NAT UNIV OF DEFENSE TECH
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202610458812.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-04-09
Publication Date
2026-06-23
Estimated Expiration
2046-04-09

AI Technical Summary

Technical Problem

Existing technologies in semi-supervised semantic segmentation of remote sensing images suffer from class imbalance due to insufficient labeled data, making it difficult to effectively utilize pseudo-label bias and hindering the model from learning features of a few classes, thus affecting segmentation accuracy and efficiency.

Method used

An initial semi-supervised semantic segmentation model is constructed. By extracting and predicting features from labeled and unlabeled data, class prototypes and weights are calculated. A bidirectional prototype loss constraint is constructed, and the model is trained by combining supervised and unsupervised losses to improve the recognition ability of minority classes.

Benefits of technology

It effectively alleviates the class imbalance problem, improves the accuracy and efficiency of semantic segmentation of remote sensing images, and enhances the model's ability to identify minority classes and the performance of semi-supervised learning.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121982576B_ABST
    Figure CN121982576B_ABST
Patent Text Reader

Abstract

The application discloses a kind of remote sensing image semi-supervised semantic segmentation method, device, equipment and medium, it is related to remote sensing image processing technical field, including: by constructing initial semi-supervised semantic segmentation model, input labeled and unlabeled data are carried out feature extraction and prediction, calculate and update labeled and unlabeled category prototype, class weight is obtained by class distribution statistics, construct bidirectional prototype loss constraint.Combining supervised loss, unsupervised consistency loss and bidirectional prototype loss constraint are carried out model training, finally realize the efficient segmentation of remote sensing image effectively alleviate the class imbalance problem, improve the recognition ability of model to minority class, enhance the performance and generalization ability of semi-supervised learning, improve the precision and efficiency of remote sensing image semantic segmentation.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Small sample unsupervised field adaptive method and device for cross-domain image classification

    CN120219827A

  • Hyperspectral image segmentation method based on semi-supervised and superpixel progressive growth

    CN120299040A