Abstract :
[en] This study introduces ResNet-GLUSE, a lightweight ResNet variant enhanced with Gated Linear Unit-enhanced Squeeze-and-Excitation (GLUSE), an adaptive channel-wise attention mechanism. By integrating dynamic gating into the traditional SE framework, GLUSE improves feature recalibration while maintaining computational efficiency. Experiments on EuroSAT and PatternNet datasets confirm its effectiveness, achieving exceeding 94% and 98% accuracy, respectively. While MobileViT achieves 99% accuracy, ResNet-GLUSE offers 33× fewer parameters, 27× fewer FLOPs, 33× smaller model size (MB), ≈6× lower power consumption (W), and ≈3× faster inference time (s), making it significantly more efficient for onboard satellite deployment. Furthermore, due to its simplicity, ResNet-GLUSE can be easily mimicked for neuromorphic computing, enabling ultra-low power inference at just 852.30 mW on Akida Brainchip. This balance between high accuracy and ultra-low resource consumption establishes ResNet-GLUSE as a practical solution for real-time Earth Observation (EO) tasks. Reproducible codes are available in our shared repository. Impact Statement-ResNet-GLUSE adds a lightweight, dynamic gated attention to a small ResNet, reaching high accuracy with a fraction of the computation, memory, and power required by existing models, enabling real-time image analytics directly on satellite edge hardware. This synergy of performance, scalability, and energy efficiency accelerates rapid decision-making in resource-constrained orbital environments, aiding critical tasks like near-real-time hazard detection and precision agriculture. Reproducible codes promote broad adoption and facilitate ongoing innovation, underscoring ResNet-GLUSE's potential as a transformative solution for next-generation EO missions.
