Related Experiment Video
Updated: Jun 21, 2026
15:25
Tomato Analyzer: A Useful Software Application to Collect Accurate and Detailed Morphological and Colorimetric Data from Two-dimensional Objects
Published on: March 16, 2010
Tomato leaf disease and severity prediction using multi-task learning
Anusri Kadam1, Parnika Jain1, Srishti Tripathi1
1Symbiosis Institute of Technology, Pune Campus, Symbiosis International (Deemed University), Pune, India.
BMC Plant Biology
|June 19, 2026
Summary
This study introduces TomatoMTL, a deep learning model for classifying tomato plant diseases and estimating their severity simultaneously. The model achieves high accuracy in both tasks, improving precision agriculture.
Area of Science:
- Agricultural Science
- Computer Science
- Plant Pathology
Background:
- Accurate plant disease identification and severity assessment are crucial for crop management and reducing agricultural losses.
- Deep learning models excel at disease classification but often neglect severity estimation, hindering informed decision-making.
Purpose of the Study:
- To develop a unified multi-task learning framework (TomatoMTL) for simultaneous tomato leaf disease classification and severity estimation.
- To enhance severity estimation accuracy through cross-task attention and feature refinement.
Main Methods:
- A ResNet50 backbone with CBAM feature refinement and task-specific branches for classification and severity prediction.
- A cross-task attention mechanism to integrate disease and severity features.
- A masking strategy for training with partially labeled data.
Main Results:
- Achieved 97.85% accuracy for disease classification and 77.66% for severity estimation.
- Outperformed state-of-the-art single-task and multi-task learning models.
- Localization analysis confirmed attention maps focus on relevant disease regions (89.4% Pointing Game accuracy).
Conclusions:
- TomatoMTL offers an effective, integrated approach for plant disease analysis.
- The framework shows strong potential for real-world precision agriculture applications.