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

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