Fungal infections: Classification performance and detectability with machine learning models

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Fungal diseases pose a significant threat to tomato crop yields.
• Accurate and early disease detection is crucial for effective crop management.
• Traditional diagnostic methods can be time-consuming and less accurate.

Purpose of the Study:

• To investigate the efficacy of AI-driven machine learning algorithms for early detection of fungal diseases in tomato plants.
• To compare the performance of various machine learning models in disease identification.
• To assess the potential for deploying these AI models in real-world agricultural settings.

Main Methods:

• Collected approximately 20,000 annotated images from two tomato cultivars (Dongfeng-1 and Gypsy) in greenhouse and open-field plots.
• Utilized high-resolution cameras, multispectral sensors, and microclimate sensors for data acquisition.
• Tested five machine learning models: CNNs, random forests, gradient boosting, SVMs, and KNN.

Main Results:

• Convolutional Neural Networks (CNNs) achieved superior diagnostic accuracy: 95.2% (±0.3) in greenhouses and 92.5% (±0.5) in open fields.
• CNNs demonstrated high AUC-ROC values (0.96 and 0.93) with low false positive rates (4.1% and 5.3%).
• AI-driven detection significantly improved accuracy and reduced fungicide application compared to visual inspection.

Conclusions:

• AI-based machine learning, particularly CNNs, offers a highly accurate and efficient method for early fungal disease detection in tomato plants.
• The developed models are robust under varying environmental conditions and feasible for deployment on edge and cloud systems.
• This technology enhances crop protection strategies and promotes sustainable agriculture.