Semi-supervised remote sensing image change detection method and system based on neighborhood relation propagation

By introducing a neighborhood relationship propagation module into semi-supervised remote sensing image change detection, and using a teacher-student framework and relevance weight matrix for weighted aggregation, the problem of ignoring pixel neighborhood correlation in existing methods is solved, achieving higher accuracy and robustness in change detection.

CN122289944AActive Publication Date: 2026-06-26江苏源驶科技有限公司
4 Cites 1 Cited by

Patent Information

Application Number
CN202610737476.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-05-27
Publication Date
2026-06-26
Estimated Expiration
2046-05-27

AI Technical Summary

Technical Problem

Existing semi-supervised remote sensing image change detection methods ignore the spatial dependence between pixels, resulting in limited detection accuracy, poor spatial consistency, and failure to effectively capture the correlation between pixels and their neighboring pixels in terms of change patterns, leading to discontinuous detection results, blurred boundaries, and a high false alarm rate.

Method used

A semi-supervised remote sensing image change detection method based on neighborhood relationship propagation is adopted. Through a teacher-student framework, multi-scale semantic features are extracted using a shared encoder and feature decoder. The squared Euclidean distance between the center pixel and its neighboring pixels is calculated by the neighborhood relationship propagation module to generate a corrected prediction map. The map is then converted into a relevance weight matrix by the Softmax function for weighted aggregation. The model is trained by combining supervised loss and unsupervised consistency loss.

Benefits of technology

It enhances the spatial consistency of the change region, obtains more continuous and clearly defined change detection results, reduces the workload of manual annotation, and maintains good detection performance under the condition of limited labeled samples. It has stronger generalization ability and robustness, especially in the case of scarce labeled scenarios.

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

Abstract

The invention discloses a semi-supervised remote sensing image change detection method and system based on neighborhood relation propagation, and belongs to the technical field of computer vision and remote sensing image processing. Comprising the following steps: acquiring a dual-temporal remote sensing image pair, and dividing a labeled data set and a non-labeled data set; constructing a semi-supervised model of a teacher-student framework, calculating a square Euclidean distance between a center pixel and a neighborhood pixel, converting the square Euclidean distance into a correlation weight matrix through Softmax, and performing neighborhood weighted aggregation on the initial prediction map to obtain a corrected prediction map; using labeled samples to calculate supervision loss, using unlabeled samples to generate pseudo labels through weak disturbance, using strong disturbance to carry out consistency learning to calculate unsupervised loss, jointly optimizing student model parameters, and updating a teacher model by exponential moving average; and inputting the double-time-phase image into the trained student model, outputting a probability graph and generating a binary change detection mask. According to the method, the spatial consistency is enhanced through neighborhood relation propagation, and the detection performance is improved when the annotation is limited.
Need to check novelty before this filing date? Find Prior Art