DDF: A Novel Dual-Domain Image Fusion Strategy for Remote Sensing Image Semantic Segmentation with Unsupervised Domain Adaptation

Lingyan Ran,Lushuang Wang,Tao Zhuo,Yinghui Xing
2024-10-24
Abstract:Semantic segmentation of remote sensing images is a challenging and hot issue due to the large amount of unlabeled data. Unsupervised domain adaptation (UDA) has proven to be advantageous in incorporating unclassified information from the target domain. However, independently fine-tuning UDA models on the source and target domains has a limited effect on the outcome. This paper proposes a hybrid training strategy as well as a novel dual-domain image fusion strategy that effectively utilizes the original image, transformation image, and intermediate domain information. Moreover, to enhance the precision of pseudo-labels, we present a pseudo-label region-specific weight strategy. The efficacy of our approach is substantiated by extensive benchmark experiments and ablation studies conducted on the ISPRS Vaihingen and Potsdam datasets.
Computer Vision and Pattern Recognition
What problem does this paper attempt to address?
The main problem that this paper attempts to solve is the Unsupervised Domain Adaptation (UDA) problem in semantic segmentation of remote sensing images. Specifically, researchers are faced with the following challenges: 1. **Effective utilization of large - scale unlabeled data**: Remote sensing images usually have a large amount of unlabeled data. How to effectively use these data for model training is a key issue. 2. **Performance degradation caused by differences between domains**: The differences between the source domain (labeled data) and the target domain (unlabeled data), such as geographical location, time change, seasonal fluctuation, etc., will lead to a significant decrease in the performance of the model on the target domain. 3. **Quality improvement of pseudo - labels**: In unsupervised domain adaptation, generating high - quality pseudo - labels is crucial for improving model performance. How to generate and use high - quality pseudo - labels is a difficult point. To address these challenges, the paper proposes a new Dual - Domain Image Fusion strategy (DDF) and a Pseudo - Label Region Weighting strategy (PRW), combining the self - training framework and the Generative Adversarial Network (GAN) method to reduce the differences between domains and improve the performance of the model on the target domain.