How Convolutional Neural Networks Diagnose Plant Disease

Area of Science:

• Plant Pathology
• Computer Science
• Artificial Intelligence

Background:

• Deep learning, particularly convolutional neural networks (CNNs), excels at plant disease classification.
• The inference process of these models remains largely a "black box," hindering reliability assessment.

Purpose of the Study:

• To interpret the decision-making process of CNNs in plant disease diagnosis.
• To validate model authenticity and training data through human-interpretable features.

Main Methods:

• Applied neuron-wise and layer-wise visualization techniques to a trained CNN.
• Utilized a publicly available plant disease image dataset.
• Optimized visualization methods and interpreted attention maps.

Main Results:

• CNNs learn disease-specific color and texture features, mirroring human diagnostic approaches.
• Identified and removed non-contributing layers, reducing parameters by 75% without accuracy loss.

Conclusions:

• Interpretable CNNs enhance the reliability and efficiency of deep learning in plant disease diagnosis.
• Findings encourage wider adoption and understanding of AI in plant science.