What question did this study set out to answer?

The aim is to develop an efficient, interpretable system for real-time plant disease detection that addresses class imbalance.

January 26, 2026PLoS ONEOpen Access

LeafAI: Interpretable plant disease detection for edge computing

View Full Paper

Ask AI

Bookmark

Authors

This hybrid AI approach enhances disease detection efficiency in agriculture, suggesting a scalable solution for farming.

The aim is to develop an efficient, interpretable system for real-time plant disease detection that addresses class imbalance.
Implemented a two-stage hybrid AI approach combining traditional machine learning and deep learning.
First stage uses a lightweight model for quick binary classification of healthy leaves.
Second stage employs deep learning models (ResNet, DenseNet, MobileNet, EfficientNet) for disease-specific classification.
Utilized Explainable AI methods like Grad-CAM to visualize model predictions.
Achieved 77.6% faster inference compared to conventional deep learning models with only 3% accuracy loss.
Reduced inference time from 4,548 seconds to 1,010.13 seconds on a large dataset of 1,227 images.
Maintained minimal CPU load while improving classification accuracy.

Kafi et al. (2026) studied this question.

View Full Paper

Ask AI

Bookmark

January 26, 2026PLoS ONEOpen Access

View Full Paper

Ask AI

Bookmark

Authors

This hybrid AI approach enhances disease detection efficiency in agriculture, suggesting a scalable solution for farming.

The aim is to develop an efficient, interpretable system for real-time plant disease detection that addresses class imbalance.
Implemented a two-stage hybrid AI approach combining traditional machine learning and deep learning.
First stage uses a lightweight model for quick binary classification of healthy leaves.
Second stage employs deep learning models (ResNet, DenseNet, MobileNet, EfficientNet) for disease-specific classification.
Utilized Explainable AI methods like Grad-CAM to visualize model predictions.
Achieved 77.6% faster inference compared to conventional deep learning models with only 3% accuracy loss.
Reduced inference time from 4,548 seconds to 1,010.13 seconds on a large dataset of 1,227 images.
Maintained minimal CPU load while improving classification accuracy.

Kafi et al. (2026) studied this question.

View Full Paper

Ask AI

Bookmark