TY - JOUR
T1 - Self-attention and frequency-augmentation for unsupervised domain adaptation in satellite image-based time series classification
AU - Gackstetter, David
AU - Yu, Kang
AU - Körner, Marco
N1 - Publisher Copyright:
© 2025 The Authors
PY - 2025/6
Y1 - 2025/6
N2 - With the increasing availability of Earth observation data in recent years, the development of deep learning algorithms for the classification of satellite image time series (SITS) has substantially progressed. Yet, when encountering settings of lacking target labels and distinct feature variations, even the latest classification algorithms may perform poorly in transferring knowledge from a trained dataset to an unknown target dataset, despite similar or even identical label sets. The research field of unsupervised domain adaptation (UDA) focuses on methods to overcome these challenges by providing algorithms that explicitly learn domain shifts between different data domains in the absence of target-labeled data. Building upon recent advances on generic UDA research in time series settings, we propose RAINCOAT-SRS, an enhancement of the frequency-augmented UDA-algorithm RAINCOAT specifically for the SITS domain. To evaluate the default and adjusted model variants, we designed several closed-label set, cross-regional and multi-temporal crop type mapping experiments, which represent common sub-problems of UDA in SITS. We first benchmark RAINCOAT against TimeMatch as a leading algorithm in this application context. Subsequently, we explored different encoder-to-decoder constellations as architectural enhancements. These analyses revealed that a combination of an self-attention-based encoder with the default decoder yields a performance increase to the standard algorithm of up to 6 % in average f1-score, and to TimeMatch by up to 24 %. Beyond, we assessed the impact of the frequency feature and SITS-specific feature extensions by integrating weather data, which both showed to improve classification accuracy only in individual sub-experiments however not consistently across the entire scope of scenarios. Finally, we outline key factors influencing the transferability, thereby emphasizing the major importance of domain-overarching stability of class-relative, structural patterns rather than of collective, linear shifts between domains. Through this research, we introduce RAINCOAT-SRS, a novel model for UDA in SITS, designed to advance generalization in remote sensing by enabling more comprehensive cross-regional and multi-temporal SITS experiments in face of lacking target-labeled data.
AB - With the increasing availability of Earth observation data in recent years, the development of deep learning algorithms for the classification of satellite image time series (SITS) has substantially progressed. Yet, when encountering settings of lacking target labels and distinct feature variations, even the latest classification algorithms may perform poorly in transferring knowledge from a trained dataset to an unknown target dataset, despite similar or even identical label sets. The research field of unsupervised domain adaptation (UDA) focuses on methods to overcome these challenges by providing algorithms that explicitly learn domain shifts between different data domains in the absence of target-labeled data. Building upon recent advances on generic UDA research in time series settings, we propose RAINCOAT-SRS, an enhancement of the frequency-augmented UDA-algorithm RAINCOAT specifically for the SITS domain. To evaluate the default and adjusted model variants, we designed several closed-label set, cross-regional and multi-temporal crop type mapping experiments, which represent common sub-problems of UDA in SITS. We first benchmark RAINCOAT against TimeMatch as a leading algorithm in this application context. Subsequently, we explored different encoder-to-decoder constellations as architectural enhancements. These analyses revealed that a combination of an self-attention-based encoder with the default decoder yields a performance increase to the standard algorithm of up to 6 % in average f1-score, and to TimeMatch by up to 24 %. Beyond, we assessed the impact of the frequency feature and SITS-specific feature extensions by integrating weather data, which both showed to improve classification accuracy only in individual sub-experiments however not consistently across the entire scope of scenarios. Finally, we outline key factors influencing the transferability, thereby emphasizing the major importance of domain-overarching stability of class-relative, structural patterns rather than of collective, linear shifts between domains. Through this research, we introduce RAINCOAT-SRS, a novel model for UDA in SITS, designed to advance generalization in remote sensing by enabling more comprehensive cross-regional and multi-temporal SITS experiments in face of lacking target-labeled data.
KW - Crop type mapping
KW - Deep learning
KW - Satellite image time series
KW - Transfer learning
KW - Unsupervised domain adaptation
UR - http://www.scopus.com/inward/record.url?scp=105002020195&partnerID=8YFLogxK
U2 - 10.1016/j.isprsjprs.2025.03.024
DO - 10.1016/j.isprsjprs.2025.03.024
M3 - Article
AN - SCOPUS:105002020195
SN - 0924-2716
VL - 224
SP - 113
EP - 132
JO - ISPRS Journal of Photogrammetry and Remote Sensing
JF - ISPRS Journal of Photogrammetry and Remote Sensing
ER -