Efficient Classification for Intelligent and Robust Satellite Image Encryption

Small Satellites for Earth Observation (SSEO) require efficient and secure image processing under strict onboard resource constraints. Traditional encryption strategies, which indiscriminately encrypt all captured images, result in excessive computational overhead and hinder real-time performance. To address this limitation, we propose a dual-stage framework that combines fast, content-aware image classification with high-performance selective encryption. First, we introduce an optimized Capsule Network (CapsNet) architecture, Faster CapsNet for Faster Classification (FCFC), to classify images according to their information sensitivity. This enables selective encryption by identifying only the sensitive images that require protection. Second, we develop a fast cryptosystem based on an advanced Galois-based key generator coupled with a reinforced dynamic bitwise encryption function, to ensure robust security with low computational complexity. Experiments conducted on a large-scale dataset of satellite images demonstrate that our approach achieves high classification accuracy (93.1%) and significantly reduces runtime compared to baseline architectures. Real-time feasibility is validated through FPGA implementation, where the cryptosystem achieves a throughput of 0.8 Gbps. The proposed framework enables intelligent, efficient, and robust protection of satellite images in resource-constrained onboard environments.