TY - GEN
T1 - Universal Domain Adaptation without Source Data for Remote Sensing Image Scene Classification
AU - Xu, Qingsong
AU - Shi, Yilei
AU - Zhu, Xiaoxiang
N1 - Publisher Copyright:
© 2022 IEEE.
PY - 2022
Y1 - 2022
N2 - Existing domain adaptation (DA) approaches are usually not well suited for practical DA scenarios of remote sensing image classification, since these methods (such as unsupervised DA) rely on rich prior knowledge about the relationship between label sets of source and target domains, and source data are usually not accessible in many cases due to the privacy or confidentiality issues. To this end, we propose a novel source data generation-based universal domain adaptation (SDG-UniDA) model, which includes two parts, i.e., the stage of source data generation and the stage of model adaptation. The first stage is to estimate the conditional distribution of source data from the pre-trained model using the knowledge of class-separability in the source domain and then to synthesize the source data. With this synthetic source data in hand, it becomes a universal DA task that requires no prior knowledge on the label sets. A novel transferable weight is proposed to distinguish the shared and private label sets to each domain, thereby promoting the adaptation in the automatically discovered shared label set and recognizing the 'unknown' samples successfully. Empirical results show that SDG-UniDA is effective and practical in this challenging setting for remote sensing image scene classification.
AB - Existing domain adaptation (DA) approaches are usually not well suited for practical DA scenarios of remote sensing image classification, since these methods (such as unsupervised DA) rely on rich prior knowledge about the relationship between label sets of source and target domains, and source data are usually not accessible in many cases due to the privacy or confidentiality issues. To this end, we propose a novel source data generation-based universal domain adaptation (SDG-UniDA) model, which includes two parts, i.e., the stage of source data generation and the stage of model adaptation. The first stage is to estimate the conditional distribution of source data from the pre-trained model using the knowledge of class-separability in the source domain and then to synthesize the source data. With this synthetic source data in hand, it becomes a universal DA task that requires no prior knowledge on the label sets. A novel transferable weight is proposed to distinguish the shared and private label sets to each domain, thereby promoting the adaptation in the automatically discovered shared label set and recognizing the 'unknown' samples successfully. Empirical results show that SDG-UniDA is effective and practical in this challenging setting for remote sensing image scene classification.
KW - Source data generation
KW - remote sensing image classification
KW - universal domain adaptation
UR - http://www.scopus.com/inward/record.url?scp=85141898140&partnerID=8YFLogxK
U2 - 10.1109/IGARSS46834.2022.9884889
DO - 10.1109/IGARSS46834.2022.9884889
M3 - Conference contribution
AN - SCOPUS:85141898140
T3 - International Geoscience and Remote Sensing Symposium (IGARSS)
SP - 5341
EP - 5344
BT - IGARSS 2022 - 2022 IEEE International Geoscience and Remote Sensing Symposium
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2022 IEEE International Geoscience and Remote Sensing Symposium, IGARSS 2022
Y2 - 17 July 2022 through 22 July 2022
ER -