DAM-Net: Domain Adaptation Network with Micro-Labeled Fine-Tuning for Change Detection

TL;DR

This work tackles the problem of poor cross-domain performance in change detection for remote sensing by introducing DAM-Net, a domain-adaptive CD framework that combines adversarial domain adaptation with a CD-specific segmentation-discriminator and a Micro-Labeled Fine-Tuning (MLFT) strategy. DAM-Net leverages a Multi-Temporal Transformer for robust multi-temporal feature fusion and a MAE-pretrained backbone, achieving superior cross-dataset results on LEVIR-CD and WHU-CD, even with labeling as low as 0.3% of the target data. The key innovations include a CD-oriented matrix-output discriminator, an alternating training strategy tailored to CD, and MLFT that selects a tiny set of target-domain samples for labeling to substantially boost adaptation, achieving performance comparable to fully supervised/semi-supervised approaches with far fewer labels. Collectively, these contributions advance practical cross-dataset CD by enabling efficient domain adaptation and offering an open-source implementation to spur further research. The approach demonstrates strong potential for scalable, label-efficient CD in diverse remote sensing contexts, with significant implications for urban monitoring, disaster response, and resource management.

Abstract

Change detection (CD) in remote sensing imagery plays a crucial role in various applications such as urban planning, damage assessment, and resource management. While deep learning approaches have significantly advanced CD performance, current methods suffer from poor domain adaptability, requiring extensive labeled data for retraining when applied to new scenarios. This limitation severely restricts their practical applications across different datasets. In this work, we propose DAM-Net: a Domain Adaptation Network with Micro-Labeled Fine-Tuning for CD. Our network introduces adversarial domain adaptation to CD for, utilizing a specially designed segmentation-discriminator and alternating training strategy to enable effective transfer between domains. Additionally, we propose a novel Micro-Labeled Fine-Tuning approach that strategically selects and labels a minimal amount of samples (less than 1%) to enhance domain adaptation. The network incorporates a Multi-Temporal Transformer for feature fusion and optimized backbone structure based on previous research. Experiments conducted on the LEVIR-CD and WHU-CD datasets demonstrate that DAM-Net significantly outperforms existing domain adaptation methods, achieving comparable performance to semi-supervised approaches that require 10% labeled data while using only 0.3% labeled samples. Our approach significantly advances cross-dataset CD applications and provides a new paradigm for efficient domain adaptation in remote sensing. The source code of DAM-Net will be made publicly available upon publication.

Figures (6)

• Figure 1: Architecture of the proposed DAM-Net.
• Figure 2: Structure of MT-Transformer.
• Figure 3: Simplified architecture of our ADA.
• Figure 4: Framework of CDMatch.
• Figure 5: (a)-(e) are results from LEVIR-CD $\rightarrow$ WHU-CD. (f)-(j) are results from WHU-CD $\rightarrow$ LEVIR-CD. (a), (b), (f), and (g) are the original images. (c) and (h) are the ground truth. The results of (d) (i) our DAM-Net w/o MLFT, (e) (j) our DAM-Net. The false positives and false negatives are indicated in red and green, respectively. Other colors represent true positives.