E-AaSSL: Hybrid EfficientNet-DeiT Framework for Ambiguity-Aware Semi-Supervised Leaf Disease Classification

DOI

10.2991/978-94-6239-707-1_10How to use a DOI?

Keywords

Semi-supervised learning; plant disease classification; ambiguity rejection; knowledge distillation; precision agriculture

Abstract

Deep learning has achieved high accuracy in plant disease classification under fully supervised settings; however, real-world agricultural applications are constrained by limited data and class imbalance. To address these challenges, this work proposes an enhanced ambiguityaware semi-supervised learning (E-AaSSL) framework tailored for plant leaf disease diagnosis, aiming to substantially reduce annotation demands while maintaining strong diagnostic performance.

The proposed pipeline integrates (i) a pretrained EfficientNet-B4 warmup phase for reliable initial pseudo-labels, (ii) ambiguity-aware pseudolabel filtering with class-balanced selection and FixMatch-style consistency regularization to suppress noise accumulation, and (iii) progressive refinement using a Vision Transformer (DeiT) through staged layer unfreezing. An ensemble of EfficientNet, a hybrid CNN–Transformer model and DeiT is further employed with an adaptive rejection mechanism to improve prediction reliability.

Experimental assessments on plant leaf disease datasets demonstrate that the framework achieves high accuracy and F1 scores under severe label scarcity, while maintaining strong selection accuracy and effective rejection of ambiguous samples. These results show that the proposed approach offers a scalable, practical solution for precision agriculture, enabling real-time crop health monitoring and minimizing economic losses in resource-constrained farming environments.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Chetan Bhatkar
AU  - Deepak D. Kshirsagar
PY  - 2026
DA  - 2026/06/18
TI  - E-AaSSL: Hybrid EfficientNet-DeiT Framework for Ambiguity-Aware Semi-Supervised Leaf Disease Classification
BT  - Proceedings of the International Conference on Recent Advances in Intelligent and Sustainable Technologies (RAIST 2026)
PB  - Atlantis Press
SP  - 110
EP  - 121
SN  - 2589-4919
UR  - https://doi.org/10.2991/978-94-6239-707-1_10
DO  - 10.2991/978-94-6239-707-1_10
ID  - Bhatkar2026
ER  -