Advancing plant disease classification using an attention-based CNN for intra-dataset and cross- dataset training

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Accurate plant disease classification is vital for global food security and agricultural output.
• Deep learning has advanced plant disease detection, but cross-dataset generalization remains a challenge.
• Few models effectively integrate intra-dataset and cross-dataset training strategies.

Purpose of the Study:

• To propose a novel attention-based Convolutional Neural Network (CNN) model.
• To enhance feature extraction and classification accuracy for plant diseases across multiple datasets.
• To address limitations in current cross-dataset training approaches for plant disease identification.

Main Methods:

• Developed an attention-based CNN model for improved feature extraction.
• Evaluated the model on five diverse datasets covering corn and potato leaf diseases.
• Implemented and compared intra-dataset and cross-dataset training methodologies.

Main Results:

• Achieved 99.38% intra-dataset classification accuracy on potato leaf images (PlantVillage dataset).
• Reached 82.93% average cross-dataset accuracy for corn leaf diseases (CD&S dataset, background removed).
• Demonstrated superior performance compared to existing techniques under similar experimental conditions.

Conclusions:

• The proposed model offers flexibility for both intra- and cross-dataset plant disease classification.
• The model's generalization capability is valuable for real-world agricultural applications with varying image quality.
• This research advances precision farming and disease management strategies through improved automated detection.