Deep Domain Adaptation Applied to Remote Sensing Data Analysis

Deep learning has transformed remote sensing, driving state-of-the-art results in land use and land cover classification, object detection, change detection, and ecological assessment. Yet a core challenge remains: models trained on data from one domain — defined by sensor type, geographic region, acquisition date, or atmospheric conditions — often degrade sharply when applied to another, even for the same task. Generating sufficient labelled reference data to retrain for each new domain is costly and time-consuming.Domain adaptation addresses this gap by enabling models to transfer knowledge across spatial, temporal, and sensor domains, with or without target-domain labels. Recent advances in foundation models and vision-language models (VLMs) are opening new frontiers in this space: large-scale pre-trained models offer rich, generalizable representations that can dramatically reduce the labelled data burden, while VLMs introduce the possibility of language-guided adaptation and zero-shot transfer to unseen remote sensing domains.This Research Topic welcomes contributions on the methodological development, evaluation, and application of deep domain adaptation for remote sensing, including:Land use and land cover mapping through domain-invariant representationsCross-sensor and cross-regional adaptation for improved classification accuracySemi-supervised and unsupervised deep adaptation techniquesFoundation model fine-tuning and prompting strategies for remote sensing domainsVision-language models for zero-shot and few-shot domain transferBenchmark datasets and evaluation protocols for domain adaptationTransfer learning and continual learning for dynamic remote sensing dataApplications in environmental monitoring, agriculture, urban analysis, and disaster response Topic editors: Gilson A O P Costa, Guilherme Lucio Abelha Mota, Pedro Juan Soto Vega Source: https://www.frontiersin.org/research-topics/80926/deep-domain-adaptation-applied-to-remote-sensing-data-analysis