The Eyes of the Gods: A Survey of Unsupervised Domain Adaptation Methods Based on Remote Sensing Data

Abstract

With the rapid development of the remote sensing monitoring and computer vision technology, the deep learning method has made a great progress to achieve applications such as earth observation, climate change and even space exploration. However, the model trained on existing data cannot be directly used to handle the new remote sensing data, and labeling the new data is also time-consuming and labor-intensive. Unsupervised Domain Adaptation (UDA) is one of the solutions to the aforementioned problems of labeled data defined as the source domain and unlabeled data as the target domain, i.e., its essential purpose is to obtain a well-trained model and tackle the problem of data distribution discrepancy defined as the domain shift between the source and target domain. There are a lot of reviews that have elaborated on UDA methods based on natural data, but few of these studies take into consideration thorough remote sensing applications and contributions. Thus, in this paper, in order to explore the further progress and development of UDA methods in remote sensing, based on the analysis of the causes of domain shift, a comprehensive review is provided with a fine-grained taxonomy of UDA methods applied for remote sensing data, which includes Generative training, Adversarial training, Self-training and Hybrid training methods, to better assist scholars in understanding remote sensing data and further advance the development of methods. Moreover, remote sensing applications are introduced by a thorough dataset analysis. Meanwhile, we sort out definitions and methodology introductions of partial, open-set and multi-domain UDA, which are more pertinent to real-world remote sensing applications. We can draw the conclusion that UDA methods in the field of remote sensing data are carried out later than those applied in natural images, and due to the domain gap caused by appearance differences, most of methods focus on how to use generative training (GT) methods to improve the model’s performance. Finally, we describe the potential deficiencies and further in-depth insights of UDA in the field of remote sensing.

Continual domain adaptation on aerial images under gradually degrading weather

“…Although DA has been extensively studied for ground-level imagery, very few works have explored DA on aerial images. Nagananda et al 14 and Xu et al 28 evaluated the state-of-the-art standard (non-continual) DA methods on aerial datasets. Nagananda et al 14 created three pairs of aerial datasets for DA based on common class labels.…”

“…Nagananda et al 14 created three pairs of aerial datasets for DA based on common class labels. However, both works 14,28 dealt with standard and non-continual DA settings, with sudden and drastic domain shifts between the source and the target domains, and did not consider gradually varying domains. To the best of our knowledge, continual DA has not yet been studied within the scope of remote sensing datasets, and there are no aerial datasets that could be utilized to assess continual DA on gradually changing environments.…”

“…Nagananda et al 14 . and Xu et al 28 . evaluated the state-of-the-art standard (non-continual) DA methods on aerial datasets.…”

Learning General-Purpose Representations for Cross-Domain Hyperspectral Images Classification with Small Samples

“…Although there have been extensive researches on UDA semantic segmentation of RSIs, the existing methods generally follow an ideal assumption that labeled source domains and unlabeled target domains have exactly the same classes [9], which can be referred to as class symmetry. As shown in Fig.…”

Urban surface classification using self-supervised domain adaptive deep learning models and its application in urban environment studies

“…It is more labor-intensive to build training datasets manually for pixel-level classi cation tasks, which is also called semantic segmentation. Furthermore, conventional deep learning models like CNNs have poor generalization ability, which means that the model is highly adapted to the training samples (source domain) and can have a signi cant performance drop when directly applied to new tasks (target domain) (Xu et al 2022). The main reason is that there are data discrepancies between the training samples and the target image.…”