Related Experiment Video
Updated: Jun 30, 2026
08:20
Superior Auto-Identification of Trypanosome Parasites by Using a Hybrid Deep-Learning Model
Published on: October 27, 2023
Swin-SHARP: a novel approach to wheat disease classification using boosted MAML and weighted ensembling with deep
Waqar Khalid1, Yazeed Alkhrijah2, Shehzad Khalid3,4
1Department of Computer Engineering, Bahria School of Engineering and Applied Sciences, Bahria University, Islamabad, Pakistan.
Scientific Reports
|June 19, 2026
Summary
A new Swin-Streamlined High Accuracy and Reduced Parameters (Swin-SHARP) transformer model significantly improves wheat disease classification accuracy. This lightweight model offers efficient feature extraction for early detection and better crop management.
Area of Science:
- Agricultural Science
- Computer Science
- Plant Pathology
Background:
- Global food security is threatened by wheat diseases, necessitating accurate early detection systems.
- Current deep learning models for wheat disease classification are computationally intensive, hindering efficient feature extraction.
- Existing methods struggle with resource constraints, limiting their application in real-time agricultural settings.
Purpose of the Study:
- To develop a lightweight and optimized transformer model for accurate and efficient wheat disease classification.
- To address the computational overhead and feature extraction limitations of existing models.
- To enhance early disease detection and improve wheat yield through advanced computational methods.
Main Methods:
- Proposed a customized, lightweight Swin-Streamlined High Accuracy and Reduced Parameters (Swin-SHARP) transformer model.
- Integrated Swin-SHARP with boosted Model-Agnostic Meta-Learning (MAML) and a weighted ensembling strategy.
- Utilized a dataset of 10,000 images encompassing various wheat diseases (brown rust, yellow rust, powdery mildew, loose smut) and healthy plants.
Main Results:
- The Swin-SHARP transformer achieved an 82.5% reduction in parameters (48.9M to 8.5M).
- The proposed model attained a remarkable 98.1% accuracy on the primary dataset, outperforming existing CNN, ensemble, and transformer models.
- Achieved 95.57% accuracy on an unseen dataset and outperformed state-of-the-art models by up to 5.8%.
Conclusions:
- The Swin-SHARP transformer offers an effective and practical solution for wheat disease classification, especially in resource-constrained environments.
- The model's efficiency and high accuracy support real-time agricultural applications for early disease detection and crop management.
- This research provides a significant advancement in leveraging AI for sustainable agriculture and enhanced food security.