Sustainable Edge Vision: An Energy-efficient Deep Learning for Crop Disease Detection

This research presents a two-stage model compression framework that combines structured channel pruning, safe pointwise incremental pruning, and knowledge distillation to optimize MobileNetV2 for sustainable edge-based crop disease detection. The proposed approach achieves 99.43% test accuracy while reducing model size by 73.84%, parameters by 23.53%, MACs by 21.86%, CPU inference latency by 26.39%, and inference energy and estimated CO₂ emissions by 28.43% compared with the baseline.