LeafFusionNet: a hybrid deep learning approach for robust plant disease detection

Area of Science:

• Agricultural Science
• Computer Vision
• Deep Learning

Background:

• Conventional crop disease diagnosis is slow and prone to human error.
• Accurate and interpretable plant disease detection is vital for sustainable agriculture.
• Deep learning and computer vision offer efficient automatic detection solutions.

Purpose of the Study:

• To introduce LeafFusionNet, a hybrid deep learning model for effective plant leaf disease identification and categorization.
• To enhance disease detection by integrating Convolutional Neural Network (CNN), Vision Transformer (ViT), and a novel attention module (LeafTAM).
• To improve the extraction of relevant texture features using Gabor filters.

Main Methods:

• Developed LeafFusionNet, a hybrid model combining CNN and ViT architectures.
• Incorporated Leaf Texture Attention Module (LeafTAM) for capturing global and local image information.
• Integrated Gabor filters to enhance the extraction of plant leaf texture features.

Main Results:

• Achieved 99.33% accuracy, 99% precision, recall, and F1 score on the Plant Village dataset.
• Demonstrated strong generalization capabilities on unseen data compared to state-of-the-art models.
• Validated the model's potential for developing robust agricultural diagnostic systems.

Conclusions:

• The proposed LeafFusionNet model offers a powerful and interpretable solution for plant disease detection.
• Hybrid deep learning models integrating CNN, ViT, and attention mechanisms show significant promise.
• Future development can leverage transformer-based vision modules and explainability techniques for enhanced agricultural diagnostics.