Clustered Personalized Federated Crop Mapping for Sentinel-2 Crop-Type Time Series

DOI

10.2991/978-94-6239-697-5_14How to use a DOI?

Keywords

agricultural monitoring; client clustering; crop-type mapping; federated learning; personalized federated learning; remote sensing; Sentinel-2; time series classification

Abstract

Federated Learning (FL), which allows the joint model training on distributed agricultural data sources without compromising data privacy, is a promising technology. However, the performance of FL is negatively impacted by high non-IID data heterogeneity, which is caused by the uneven distribution of various crops, management practices, and environmental conditions in different regions. In this paper, we explore the problem of federated crop-type classification on the TimeSen2Crop dataset, which is a large Sentinel-2 pixel-level benchmark with over one million samples on 16 crop types. The performance of the model varies greatly when the standard FedAvg algorithm is directly applied to the data. To mitigate the issue of non-IID data heterogeneity, we propose a framework called “clustered personalized federated learning.” In the proposed framework, the clients are grouped based on their privacy-preserving signatures, which are computed from the label distribution histograms and the class-conditional feature prototypes learned from a lightweight temporal encoder. Finally, the federated optimization is performed on each cluster with a shared backbone network and adaptive heads. In the experiment, we simulate 100 clients and find that the proposed framework improves the macro-F1 and worst client macro-F1 compared to the standard FedAvg algorithm and local training.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Durganand Jha
AU  - Amandeep Kaur
PY  - 2026
DA  - 2026/06/04
TI  - Clustered Personalized Federated Crop Mapping for Sentinel-2 Crop-Type Time Series
BT  - Proceedings of the Conference on Bridging Engineering Disciplines with AI and Machine Learning (BEDAIML 2026)
PB  - Atlantis Press
SP  - 150
EP  - 164
SN  - 1951-6851
UR  - https://doi.org/10.2991/978-94-6239-697-5_14
DO  - 10.2991/978-94-6239-697-5_14
ID  - Jha2026
ER  -