AdaSemiCD: An Adaptive Semi-Supervised Change Detection Method Based on Pseudo-Label Evaluation

TL;DR

AdaSemiCD addresses the challenge of scarce labeled data in bi-temporal remote sensing change detection by introducing an adaptive semi-supervised framework that quality-controls pseudo-labels. It defines an entropy-based pseudo-label qualification metric with class-balancing and regional amplification, and integrates AdaFusion for adaptive region/content fusion and AdaEMA for data-aware EMA updates. Empirical results on ten CD datasets demonstrate state-of-the-art or competitive performance across various partitions, with notable gains on IoU$^c$ and maintained overall accuracy, highlighting improved robustness to noisy pseudo-labels. The proposed adaptive strategy reduces labeling burdens while enhancing training stability, offering potential applicability to other semi-supervised tasks in remote sensing and beyond.

Abstract

Change Detection (CD) is an essential field in remote sensing, with a primary focus on identifying areas of change in bi-temporal image pairs captured at varying intervals of the same region by a satellite. The data annotation process for the CD task is both time-consuming and labor-intensive. To make better use of the scarce labeled data and abundant unlabeled data, we present an adaptive dynamic semi-supervised learning method, AdaSemiCD, to improve the use of pseudo-labels and optimize the training process. Initially, due to the extreme class imbalance inherent in CD, the model is more inclined to focus on the background class, and it is easy to confuse the boundary of the target object. Considering these two points, we develop a measurable evaluation metric for pseudo-labels that enhances the representation of information entropy by class rebalancing and amplification of confusing areas to give a larger weight to prospects change objects. Subsequently, to enhance the reliability of sample-wise pseudo-labels, we introduce the AdaFusion module, which is capable of dynamically identifying the most uncertain region and substituting it with more trustworthy content. Lastly, to ensure better training stability, we introduce the AdaEMA module, which updates the teacher model using only batches of trusted samples. Experimental results from LEVIR-CD, WHU-CD, and CDD datasets validate the efficacy and universality of our proposed adaptive training framework.

Figures (6)

• Figure 1: The overview of the proposed AdaSemiCD framework and details of two adaptive modules. Our approach is based on the common MT training pipeline for SSCD, and we suggest utilizing the AdaFusion module to produce samples with increased reliability and the AdaEMA module to enhance the efficiency of EMA update.
• Figure 2: Class statistics of common CD datasets. The proportion of the changed/unchanged categories is extremely unbalanced.
• Figure 3: Typical examples of ten change detection datasets we use, with Pre-event image, Post-event image, and Ground Truth.
• Figure 4: Visualizations of different models on six building change detection datasets at the 5$\%$ labeled training ratio.
• Figure 5: Visualizations of different models on four mutil-class change detection datasets at the 5$\%$ labeled training ratio.