SynerFANet: a synergistic hybrid architecture for advanced plant leaf disease detection

Accurate identification of plant leaf diseases is essential for modern agriculture. This paper presents SynerFANet, a hybrid deep learning framework designed to improve disease detection in sugarcane leaves. SynerFANet integrates two core modules: AdaptiveMBNet and WiseAttentionNet for comprehensive feature extraction and processing. AdaptiveMBNet combines MBConv layers with attention mechanisms to improve feature quality and reduce computation, enabling more accurate disease detection. WiseAttentionNet incorporates attention mechanisms into depthwise and expansion layers to enhance feature recalibration. The combination of the two cores inside SynerFANet can improve the representation capacity and robustness of the overall model. We evaluate the model using two datasets: a new proposed field-collected SugarLeaf-IDN dataset and the publicly available PlantVillage dataset. SynerFANet achieves superior accuracy with a moderate parameter size and GFLOPs, providing a balanced trade-off between predictive performance and computational cost, and exhibiting stable convergence during training. SynerFANet achieves 95.81% validation accuracy on our challenging real-world SugarLeaf-IDN dataset and 99.85% (SOTA) on the controlled PlantVillage benchmark.