Interpretable deep multimodal-based tomato disease diagnosis and severity estimation.

Area of Science:

• Agricultural Science
• Computer Science
• Data Science

Background:

• Plant diseases, especially in vital crops like tomatoes, significantly threaten global food security.
• Traditional farming methods often lack the efficiency to combat widespread crop diseases effectively.
• Unimodal systems for disease detection have limitations in accuracy and interpretability.

Purpose of the Study:

• To introduce a novel multimodal deep learning algorithm for enhanced plant disease detection and severity prediction in tomatoes.
• To overcome the limitations of unimodal approaches by integrating visual and environmental data.
• To improve the accuracy and interpretability of plant disease classification and severity assessment.

Main Methods:

• Utilized EfficientNetB0 for image-based classification of tomato plant diseases.
• Employed Recurrent Neural Networks (RNN) for predicting disease severity from environmental data.
• Integrated visual and climatological inputs into a multimodal deep learning framework.
• Applied LIME and SHAP explainable AI techniques for outcome interpretability.

Main Results:

• Achieved 96.40% accuracy in plant disease classification.
• Reached 99.20% accuracy in predicting disease severity.
• Demonstrated enhanced classification accuracy and interpretability compared to unimodal systems.
• Provided insights into disease severity classification through explainable AI.

Conclusions:

• The multimodal deep learning model offers a significant advancement in plant disease management.
• The approach supports precision agriculture practices and strengthens local food system resilience.
• This technology has the potential to mitigate disease impacts and enhance food security, particularly for tomato-dependent economies.