Deep learning-based automatic diagnosis of rice leaf diseases using ensemble CNN models.

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Rice diseases significantly threaten global crop yields and food security.
• Traditional diagnostic methods for rice diseases are often labor-intensive, time-consuming, and require specialized expertise.
• There is a critical need for efficient, accurate, and scalable tools for early rice disease detection.

Purpose of the Study:

• To develop and evaluate a deep learning-based automated diagnostic system for identifying six common rice leaf diseases.
• To compare the performance of seven advanced deep learning architectures for rice disease classification.
• To create an ensemble model for enhanced diagnostic accuracy and robustness.

Main Methods:

• A large-scale dataset of annotated rice leaf images covering six diseases was curated.
• Seven deep learning models (MobileNetV2, GoogLeNet, EfficientNet, ResNet-34, DenseNet-121, VGG16, ShuffleNetV2) were trained and evaluated.
• An ensemble model was constructed by integrating the top-performing individual networks using an average fusion strategy.

Main Results:

• GoogLeNet, DenseNet-121, ResNet-34, and VGG16 exhibited superior performance in accuracy and reduced class confusion.
• The ensemble model demonstrated significantly reduced misclassification rates compared to individual models.
• The system achieved robust and scalable diagnostic capabilities, validated on diverse environmental data.

Conclusions:

• Deep learning, particularly ensemble methods, offers a powerful approach for automated rice leaf disease diagnosis.
• The developed system provides a reliable and scalable solution for real-world agricultural applications, aiding in timely crop management.
• This technology has the potential to enhance crop productivity and mitigate losses caused by rice diseases.