Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GTPC-SSCD: Gate-guided Two-level Perturbation Consistency-based Semi-Supervised Change Detection

Yan Xing, Qi'ao Xu, Zongyu Guo, Rui Huang, Yuxiang Zhang

TL;DR

This work tackles semi-supervised change detection (SSCD) under limited labeled data by introducing GTPC-SSCD, which integrates image-level strong-to-weak consistency with feature-level perturbation consistency. A hardness-based gating mechanism selectively applies feature perturbations to more challenging samples, maximizing unlabeled data utilization. The method employs a Siamese ResNet-50 change detector with an ASPP decoder and seven perturbation types, optimized via a composite loss combining supervised and two unsupervised terms. Across six remote-sensing CD datasets, GTPC-SSCD achieves state-of-the-art performance with notable gains on WHU-CD and GZ-CD, while maintaining reasonable computational efficiency. These results demonstrate the value of adaptive, two-level perturbation strategies for leveraging unlabeled data in SSCD.

Abstract

Semi-supervised change detection (SSCD) utilizes partially labeled data and abundant unlabeled data to detect differences between multi-temporal remote sensing images. The mainstream SSCD methods based on consistency regularization have limitations. They perform perturbations mainly at a single level, restricting the utilization of unlabeled data and failing to fully tap its potential. In this paper, we introduce a novel Gate-guided Two-level Perturbation Consistency regularization-based SSCD method (GTPC-SSCD). It simultaneously maintains strong-to-weak consistency at the image level and perturbation consistency at the feature level, enhancing the utilization efficiency of unlabeled data. Moreover, we develop a hardness analysis-based gating mechanism to assess the training complexity of different samples and determine the necessity of performing feature perturbations for each sample. Through this differential treatment, the network can explore the potential of unlabeled data more efficiently. Extensive experiments conducted on six benchmark CD datasets demonstrate the superiority of our GTPC-SSCD over seven state-of-the-art methods.

GTPC-SSCD: Gate-guided Two-level Perturbation Consistency-based Semi-Supervised Change Detection

TL;DR

This work tackles semi-supervised change detection (SSCD) under limited labeled data by introducing GTPC-SSCD, which integrates image-level strong-to-weak consistency with feature-level perturbation consistency. A hardness-based gating mechanism selectively applies feature perturbations to more challenging samples, maximizing unlabeled data utilization. The method employs a Siamese ResNet-50 change detector with an ASPP decoder and seven perturbation types, optimized via a composite loss combining supervised and two unsupervised terms. Across six remote-sensing CD datasets, GTPC-SSCD achieves state-of-the-art performance with notable gains on WHU-CD and GZ-CD, while maintaining reasonable computational efficiency. These results demonstrate the value of adaptive, two-level perturbation strategies for leveraging unlabeled data in SSCD.

Abstract

Semi-supervised change detection (SSCD) utilizes partially labeled data and abundant unlabeled data to detect differences between multi-temporal remote sensing images. The mainstream SSCD methods based on consistency regularization have limitations. They perform perturbations mainly at a single level, restricting the utilization of unlabeled data and failing to fully tap its potential. In this paper, we introduce a novel Gate-guided Two-level Perturbation Consistency regularization-based SSCD method (GTPC-SSCD). It simultaneously maintains strong-to-weak consistency at the image level and perturbation consistency at the feature level, enhancing the utilization efficiency of unlabeled data. Moreover, we develop a hardness analysis-based gating mechanism to assess the training complexity of different samples and determine the necessity of performing feature perturbations for each sample. Through this differential treatment, the network can explore the potential of unlabeled data more efficiently. Extensive experiments conducted on six benchmark CD datasets demonstrate the superiority of our GTPC-SSCD over seven state-of-the-art methods.

Paper Structure

This paper contains 13 sections, 13 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Our GTPC-SSCD method
  4. Supervised training part
  5. Unsupervised training part
  6. Loss function
  7. Implementation detail
  8. Change detection network
  9. Super-parameters
  10. Experiment
  11. Experimental Setup
  12. Results and Discussion
  13. Conclusion

Figures (4)

  • Figure 1: Motivation analysis of SSCD with different perturbation variants under $5\%$ labeled training data on WHU-CD and GZ-CD datasets. Upper part: Visual cases of different variants. Here, white for TP, black for TN, red for FP, and green for FN. Lower part: Performance comparison of different variants. Sup-only: our method uses only labeled data. Feature: Feature-level perturbation consistency. Image: Image-level strong-to-weak consistency. Ours: Gate-guided two-level perturbation consistency.
  • Figure 2: Framework of the proposed GTPC-SSCD method.
  • Figure 3: Detection results of different methods on six CD datasets at $5\%$ labeled ratio. Here, white for TP, black for TN, red for FP, and green for FN.
  • Figure 4: Comparison of different perturbation ratios at $5\%$ labeled WHU-CD.