FOX-SHIELD: A Secure Encrypted-Domain Remote Sensing Image Classification Framework Using Chebyshev-SHA and Fox-Optimized FRNNs

Remote sensing (RS) images transmitted through distributed satellite, UAV, and sensor networks are vulnerable to interception and misuse, while conventional encryption may reduce the effectiveness of downstream image classification. This paper proposes FOX-SHIELD, a secure encrypted-domain remote sensing image classification framework that integrates Chebyshev-SHA dynamic encryption with a Fox-Optimized Fast Recurrent Neural Network. The Chebyshev-SHA module generates image-specific dynamic keys and applies permutation–diffusion encryption to protect image confidentiality and integrity. The encrypted image representation is then classified using an FRNN optimized by the Fox Optimization Algorithm to improve convergence, classification accuracy, and computational efficiency. Experiments on the UC Merced Land Use and NWPU-RESISC45 datasets evaluate FOX-SHIELD across different scene-classification settings. On UC Merced, FOX-SHIELD achieves 97.14% accuracy, 96.87% recall, and 96.96% F1-score, outperforming selected privacy-preserving and lightweight remote sensing classification baselines. Empirical security analysis further indicates improved entropy, key sensitivity, and resistance to statistical attacks. The results suggest that FOX-SHIELD is a promising framework for secure remote sensing image classification in distributed, resource-aware environments.