Rapid Grapevine Health Diagnosis Based on Digital Imaging and Deep Learning.

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Accurate grapevine disease detection is crucial for crop yield.
• Existing deep learning models lack practical farmer-friendly applications.
• Need for intelligent systems for early disease identification and prevention.

Purpose of the Study:

• Develop an intelligent, user-friendly software (AI GrapeCare) for grapevine disease detection.
• Utilize RGB imagery and hybrid deep learning networks for diagnosis.
• Provide a rapid tool to assist farmers in disease management.

Main Methods:

• Combined Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM), and deep neural networks (DNNs).
• Employed transfer learning (VGG16, VGG19, ResNet50, ResNet101V2) and Gray Level Co-occurrence Matrix (GLCM) for textural analysis.
• Utilized a plant disease detection (PDD) dataset, data augmentation, and a hybrid CNN_RGB-LSTM_GLCM model based on VGG16.

Main Results:

• The hybrid CNN_RGB-LSTM_GLCM model achieved 96.6% validation accuracy, precision, recall, and F-measure.
• Achieved 93.4% intersection over union and a loss of 0.123.
• The AI GrapeCare software provides diagnoses in under one minute.

Conclusions:

• The developed AI GrapeCare system offers a highly accurate and rapid solution for grapevine disease detection.
• The hybrid deep learning approach significantly outperforms individual models and non-augmented data.
• The framework has potential for expansion to other tree disease diagnostics, aiding broader agricultural applications.