Related Experiment Videos
A hybrid SE-ResNet50 deep learning framework for high-accuracy and explainable cotton leaf disease classification.
Emrah Aslan1, Yıldırım Özüpak2
1Faculty of Engineering and Architecture, Mardin Artuklu University, Mardin, Turkey.
BMC Plant Biology
|May 27, 2026
Summary
This study introduces a hybrid deep learning model for accurate cotton leaf disease identification, improving crop management and yield. The SE-ResNet50 framework achieves high accuracy and interpretability for precision agriculture.
Area of Science:
- Agricultural Science
- Computer Science
- Plant Pathology
Background:
- Cotton production faces significant losses due to foliar diseases and pests.
- Accurate and automated disease identification is crucial for sustainable crop management and maintaining fiber quality.
Purpose of the Study:
- To develop a hybrid deep learning framework for enhanced cotton leaf disease classification.
- To improve the accuracy and interpretability of automated disease identification systems.
Main Methods:
- A hybrid deep learning model integrating ResNet50 with Squeeze-and-Excitation (SE) channel attention modules was proposed.
- The model was trained on a six-class cotton leaf disease dataset and optimized using Weighted CrossEntropyLoss, Adam optimization, and learning rate scheduling.
- Grad-CAM was used for visualizing model attention and enhancing interpretability.
Main Results:
- The SE-ResNet50 model achieved high performance with 99.72% training accuracy and 99.31% validation accuracy.
- Model convergence was achieved by the 14th epoch.
- Visualization confirmed the model's focus on biologically relevant symptom regions, enhancing interpretability.
Conclusions:
- The proposed SE-ResNet50 framework provides a highly accurate and interpretable solution for cotton leaf disease monitoring.
- This approach is suitable for real-world applications in precision agriculture.
- The model outperforms existing methods in accuracy and transparency.
