Abstract:Semi-supervised change detection (SSCD) employs partially labeled data and a substantial amount of unlabeled data to identify differences between images captured in the same geographic area but at different times. However, existing consistency regularization-based SSCD methods only implement perturbations at a single level and can not exploit the full potential of unlabeled data. In this paper, we introduce a novel Gate-guided Two-level Perturbation Consistency regularization-based SSCD method (GTPC-SSCD), which simultaneously maintains strong-to-weak consistency at the image level and perturbation consistency at the feature level, thus effectively utilizing the unlabeled data. Moreover, a gate module is designed to evaluate the training complexity of different samples and determine the necessity of performing feature perturbations on each sample. This differential treatment enables the network to more effectively explore the potential of unlabeled data. Extensive experiments conducted on six public remote sensing change detection datasets demonstrate the superiority of our method over seven state-of-the-art SSCD methods.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to make more effective use of a large amount of unlabeled data to improve model performance in semi - supervised change detection (SSCD). Specifically, the existing methods based on consistency regularization only achieve perturbation at a single level and cannot fully utilize the potential of unlabeled data. ### Specific description of the problem 1. **Limitations of existing methods**: - Most of the existing SSCD methods are based on consistency regularization, but these methods usually only implement strong - to - weak consistency at the image level or perform simple perturbation consistency at the feature level. This results in insufficient utilization of unlabeled data. - The single - level perturbation method cannot comprehensively mine the information in unlabeled data, which limits the performance improvement of the model. 2. **Research objectives**: - Propose a new Gate - guided Two - level Perturbation Consistency regularization - based SSCD (GTPC - SSCD for short), which can maintain both strong - to - weak consistency at the image level and perturbation consistency at the feature level simultaneously. - Design a gate module to evaluate the training complexity of different samples and decide whether to apply feature perturbation to each sample, so as to more effectively explore the potential of unlabeled data. ### Key points of the solution - **Two - level perturbation consistency**: By introducing perturbations at both the image level and the feature level simultaneously, ensure that the model can learn more information from different perspectives, and improve the robustness and generalization ability of the model. - **Gating mechanism**: Introduce a gate module to distinguish the training difficulty of different samples, and intelligently decide whether to apply feature perturbation to a certain sample, avoid unnecessary computational overhead, and improve training efficiency. ### Experimental verification The paper verifies the effectiveness and superiority of the proposed method through extensive experiments on six publicly available remote - sensing change - detection datasets, which is significantly better than the other seven state - of - the - art SSCD methods. ### Summary This paper aims to solve the shortcomings of existing SSCD methods in using unlabeled data by introducing two - level perturbation consistency and gating mechanism, thereby improving the performance of change - detection tasks.

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