Answers are grounded in this paper and cite their sources.
Maize leaf diseases can reduce crop yield and quality, making timely detection important. However, existing detectors still face trade-offs between detection accuracy and computational efficiency, particularly when further reducing model complexity for edge deployment. MELD-YOLO is developed as a lightweight detector based on YOLO26n. The backbone convolutions are replaced with GhostConv to reduce computational cost, and the parameter-free SimAM module is introduced to enhance disease-related feature responses. Feature Channel Reconfiguration (FCR) adjusts channel widths for maize leaf disease detection. Layer-adaptive magnitude-based pruning (LAMP) then compresses the model, while feature-based distillation during fine-tuning recovers the lost accuracy. Using the Corn Leaf Diseases dataset, MELD-YOLO achieves 98.5% mAP@50 and 91.3% mAP@50-95 with 0.875 M parameters and 2.0 GFLOPs, with 63.4% fewer parameters, 61.5% less computation and a 2.4-point gain in mAP@50-95 over YOLO26n. Pruning and distillation then reduce the model to 0.505 M parameters and 1.5 GFLOPs while retaining 90.4% mAP@50-95. After quantisation, the model achieves 12.38 FPS with 81.05 ms latency and a 0.65 MB model size when running on the Raspberry Pi 5B, compared with 4.19 FPS before quantisation. These results support real-time maize leaf disease detection on resource-constrained edge devices.
No takes yet. Share an insight, caveat, or question.
Sun et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: