EUVIMEDΑνακαλύψτε το EUVIMED
Open access · CC BY-NC-ND 4.0

Cross-architectural mixture-of-experts with adaptive soft routing for plant leaf disease classification

Phi-Hung Hoang, Thi-Thu-Hong Phan

Array · 2026-12 · 32 · 101261

DOI: 10.1016/j.array.2026.101261 ↗

At a glance

Journal article in Array, 2026-12. Written by Phi-Hung Hoang and 1 co-authors. Cited 1 times so far (Crossref); 43 references. The full text is freely available (open access, CC BY-NC-ND 4.0).

ΠερίληψηΒιβλιογραφικά στοιχείαCiteRelated studies

Abstract (original language)

Plant leaf disease classification is crucial for crop protection and precision agriculture but remains challenging due to complex backgrounds, illumination variations, severe class imbalance, and high intra-class variability. While convolutional neural networks (CNNs) effectively capture local texture patterns and vision transformers provide stronger global contextual modeling, existing approaches typically rely on a single architecture or static fusion strategies that assign fixed importance to heterogeneous representations regardless of input characteristics. Consequently, they may fail to adapt to the diverse visual manifestations of plant diseases. To address these limitations, we propose an adaptive soft Mixture-of-Experts (ASMoE) framework with cross-architectural routing that integrates EfficientNet-B0, DenseNet 121, and Swin-Tiny to exploit complementary local, multi-scale, and global representations. A cross-architectural soft gating mechanism dynamically assigns input-dependent expert weights, enabling adaptive collaboration among CNN- and Transformer-based experts according to the visual characteristics of each input. In addition, a two-stage refinement strategy is employed to further refine the learned representations and improve performance. Experiments on a highly imbalanced potato leaf disease dataset achieve 91.68% recall and 92.62% F1-score, surpassing the strongest individual expert by 5.91% and 5.03%, respectively. Additional evaluations on durian and sesame leaf disease datasets yield F1-scores of 94.03% and 97.04%, demonstrating consistent effectiveness across different crops and disease distributions. Furthermore, interpretability analyses reveal complementary expert specialization and adaptive routing behavior, providing qualitative evidence that different disease patterns are associated with distinct combinations of visual representations. These findings highlight the effectiveness of adaptive cross-architectural expert collaboration for heterogeneous representation learning.

Cite

APA

Hoang, P., & Phan, T. (2026). Cross-architectural mixture-of-experts with adaptive soft routing for plant leaf disease classification. Array, 32, 101261. https://doi.org/10.1016/j.array.2026.101261

Vancouver

Hoang P, Phan T. Cross-architectural mixture-of-experts with adaptive soft routing for plant leaf disease classification. Array. 2026;32:101261. doi:10.1016/j.array.2026.101261

BibTeX
@article{Hoang2026,
  author = {Hoang, Phi-Hung and Phan, Thi-Thu-Hong},
  title = {Cross-architectural mixture-of-experts with adaptive soft routing for plant leaf disease classification},
  journal = {Array},
  year = {2026},
  volume = {32},
  pages = {101261},
  doi = {10.1016/j.array.2026.101261}
}
Start for free →