Hierarchical Federated Learning with Cross-Modal Representation Fusion and Secure Transform for Precision Target Selection in Advanced Thin Film Imaging Modalities

Authors

  • Dr.Janarthanam S Author
  • Prof. Raja Sarath Kumar Boddu Author
  • Dr.B. Vivekanadam Author
  • Dr.Shailendra Aswale Author
  • Mohmad Ahmed Ali Author
  • Amit Verma Author
  • Elangovan Muniyandy Author
  • Koppuravuri Gurnadha Gupta Author

Keywords:

Privacy-preserving learning; Thin-film imaging modalities; Precision target selection; Multimodal imaging; secure aggregation; intelligent imaging.

Abstract

The higher level of the thin-film imaging modalities produces heterogeneous and high-resolution data which provide precision target-selection and also cause serious problems regarding the data privacy, and communication efficiency, the cross-modal representation learning. In such distributed imaging environments, centralised forms of learning are also frequently not practical because of privacy limitations and due to scaling to large amounts of data. This paper presents a hierarchical federated learning model to deal with these issues which combines cross-modal representation fusion and a secure transform to meet privacy-preserving precision target selection. The offered architecture is a multi-layer design of client-edge-cloud aggregation to effectively train worldwide models over the distributed imaging nodes, and reduce non-IID heterogeneity in data. A cross modal fusion approach is to learn similarity representations of a multiplicity of complementary representations by using thin-film imaging modalities, allowing powerful and discriminative target characterization. All data confidentiality and model security conditions are ensured by a secure feature transformation and aggregation scheme that does not need data sharing in raw form. Wide experimental testing of multimodal thin-film imaging data sets indicates that the proposed methodology effectively achieves superior accuracy, precision, recall, F1 score, and mean average precision compared to centralised learning, conventional federated learning, and single modes of learning, in addition to incurring much lower communication expenses and retaining privacy. The outcomes confirm the feasibility of the suggested framework in the context of safe, scalable and high accuracy in the target selection in sophisticated imaging systems.

Downloads

Download data is not yet available.

Downloads

Published

2026-09-22

How to Cite

Hierarchical Federated Learning with Cross-Modal Representation Fusion and Secure Transform for Precision Target Selection in Advanced Thin Film Imaging Modalities. (2026). Minimax Theory and Its Applications, 11(2s), 639-649. https://journalmta.com/index.php/jmta/article/view/443