Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Free Lunch for Federated Remote Sensing Target Fine-Grained Classification: A Parameter-Efficient Framework

Shengchao Chen, Ting Shu, Huan Zhao, Jiahao Wang, Sufen Ren, Lina Yang

TL;DR

This work tackles privacy and resource constraints in remote-sensing target fine-grained classification by introducing PRFL, a privacy-preserving federated framework that combines synchronized bidirectional knowledge distillation with dynamic parameter decomposition. Each client maintains a local teacher and student model, with only the student parameters uploaded for server aggregation, while a global model is formed via FedAvg on the student weights; knowledge is transferred bidirectionally to tailor local representations. The dynamic parameter decomposition uses low-rank updates and an AIC-guided selection to minimize communication while preserving performance, making the approach practical for edge devices. Extensive experiments on MTARSI, MSTAR, FGSCR-42, and Aircraft-16 under pathological and Dirichlet non-IID settings show PRFL consistently outperforms state-of-the-art baselines, with quantified gains in accuracy and stability, even under stringent communication budgets, and differential privacy experiments demonstrate robust privacy protection with manageable accuracy trade-offs. This framework advances practical, privacy-preserving collaboration for remote-sensing TFGC in cross-regional settings, enabling robust personalization with limited bandwidth.

Abstract

Remote Sensing Target Fine-grained Classification (TFGC) is of great significance in both military and civilian fields. Due to location differences, growth in data size, and centralized server storage constraints, these data are usually stored under different databases across regions/countries. However, privacy laws and national security concerns constrain researchers from accessing these sensitive remote sensing images for further analysis. Additionally, low-resource remote sensing devices encounter challenges in terms of communication overhead and efficiency when dealing with the ever-increasing data and model scales. To solve the above challenges, this paper proposes a novel Privacy-Reserving TFGC Framework based on Federated Learning, dubbed PRFL. The proposed framework allows each client to learn global and local knowledge to enhance the local representation of private data in environments with extreme statistical heterogeneity (non. Independent and Identically Distributed, IID). Thus, it provides highly customized models to clients with differentiated data distributions. Moreover, the framework minimizes communication overhead and improves efficiency while ensuring satisfactory performance, thereby enhancing robustness and practical applicability under resource-scarce conditions. We demonstrate the effectiveness of the proposed PRFL on the classical TFGC task by leveraging four public datasets.

Free Lunch for Federated Remote Sensing Target Fine-Grained Classification: A Parameter-Efficient Framework

TL;DR

This work tackles privacy and resource constraints in remote-sensing target fine-grained classification by introducing PRFL, a privacy-preserving federated framework that combines synchronized bidirectional knowledge distillation with dynamic parameter decomposition. Each client maintains a local teacher and student model, with only the student parameters uploaded for server aggregation, while a global model is formed via FedAvg on the student weights; knowledge is transferred bidirectionally to tailor local representations. The dynamic parameter decomposition uses low-rank updates and an AIC-guided selection to minimize communication while preserving performance, making the approach practical for edge devices. Extensive experiments on MTARSI, MSTAR, FGSCR-42, and Aircraft-16 under pathological and Dirichlet non-IID settings show PRFL consistently outperforms state-of-the-art baselines, with quantified gains in accuracy and stability, even under stringent communication budgets, and differential privacy experiments demonstrate robust privacy protection with manageable accuracy trade-offs. This framework advances practical, privacy-preserving collaboration for remote-sensing TFGC in cross-regional settings, enabling robust personalization with limited bandwidth.

Abstract

Remote Sensing Target Fine-grained Classification (TFGC) is of great significance in both military and civilian fields. Due to location differences, growth in data size, and centralized server storage constraints, these data are usually stored under different databases across regions/countries. However, privacy laws and national security concerns constrain researchers from accessing these sensitive remote sensing images for further analysis. Additionally, low-resource remote sensing devices encounter challenges in terms of communication overhead and efficiency when dealing with the ever-increasing data and model scales. To solve the above challenges, this paper proposes a novel Privacy-Reserving TFGC Framework based on Federated Learning, dubbed PRFL. The proposed framework allows each client to learn global and local knowledge to enhance the local representation of private data in environments with extreme statistical heterogeneity (non. Independent and Identically Distributed, IID). Thus, it provides highly customized models to clients with differentiated data distributions. Moreover, the framework minimizes communication overhead and improves efficiency while ensuring satisfactory performance, thereby enhancing robustness and practical applicability under resource-scarce conditions. We demonstrate the effectiveness of the proposed PRFL on the classical TFGC task by leveraging four public datasets.
Paper Structure (31 sections, 14 equations, 3 figures, 12 tables, 1 algorithm)

This paper contains 31 sections, 14 equations, 3 figures, 12 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Remote Sensing Target Fine-Grained Classification
  4. Privacy-Preserving Federated Learning
  5. Preliminaries
  6. Federated Learning
  7. Problem Formulation for PRFL
  8. Methodology
  9. Framework Architecture
  10. Synchronized Bidirectional Knowledge Distillation
  11. Dynamic Parameter Decomposition
  12. Akaike Information Criterion Constraint
  13. AIC-constraint Variance Explained
  14. Experiments and Results
  15. Dataset
  16. ...and 16 more sections

Figures (3)

  • Figure 1: Schematic diagram of the proposed framework.
  • Figure 2: Partial visualization of the benchmark dataset used in this study. From top to bottom are MSTAR, Aircraft-16, MTARSI, and FGSCR-42. It is clear that the distributions of the different labeled samples in these datasets are extremely similar.
  • Figure 3: Visualization of four benchmark datasets. From top to bottom are MTARSI, Aircraft-16, MSTAR, and FGSCR-42, where a larger red circle means that the client has more such samples, and the opposite means that there are fewer such samples on the client.