sub1:assertion {
  sub1:comparatorGroup dcterms:description "Centralized training with pooled data, transfer learning from publicly available pre-trained models, independent training without collaboration, and traditional data sharing agreements" .
  sub1:federated-learning-for-satellite-image-classificat pico:comparatorGroup sub1:comparatorGroup ;
    pico:interventionGroup sub1:interventionGroup ;
    pico:outcomeGroup sub1:outcomeGroup ;
    pico:population sub1:population ;
    dcterms:description "For satellite imagery classification tasks involving multiple data holders, how do federated learning approaches compare to centralized training and independent training in achieving model accuracy while preventing raw data exposure and enabling cross-organizational collaboration?" ;
    a pico:PICO , sciencelive:EffectivenessResearchQuestions ;
    rdfs:label "Federated Learning for Satellite Image Classification: A Scoping Review" .
  sub1:interventionGroup dcterms:description "Federated learning approaches enabling collaborative model training without sharing raw imagery, including horizontal federated learning, vertical federated learning, federated transfer learning, and privacy-enhanced variants with differential privacy or secure aggregation" .
  sub1:outcomeGroup dcterms:description "Model accuracy and performance metrics (F1, IoU, mAP), communication efficiency (bandwidth requirements, convergence speed), privacy preservation (data leakage risk, gradient inversion attack resistance), scalability across heterogeneous sensors and resolutions, and practical deployment feasibility" .
  sub1:population dcterms:description "Multi-source satellite imagery datasets held by different organizations including space agencies (ESA, NASA, JAXA), commercial providers (Planet, Maxar, Airbus), national mapping agencies, and research institutions, covering applications such as land cover classification, object detection, change monitoring, and disaster response" .
}