Cross-Architectural Mixture-of-Experts with Adaptive Soft Routing for Plant Leaf Disease Classification

The study introduces an adaptive soft Mixture-of-Experts (MoE) framework for plant leaf disease classification, integrating EfficientNet-B0, DenseNet-121, and Swin-Tiny architectures to leverage multi-scale features. Utilizing a soft gating mechanism for input-dependent expert weights and a two-stage refinement training strategy, the model achieved a recall of 91.68% and an F1-score of 92.62% on a potato leaf disease dataset, outperforming individual models. This approach addresses challenges in class imbalance and representation learning, offering significant improvements for practitioners in precision agriculture and crop health monitoring.