Diagnosis of Citrus Greening Using Artificial Intelligence: A Faster Region-Based Convolutional Neural Network...

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Citrus Greening (CG), or Huanglongbing, is a devastating citrus disease transmitted by vectors.
• Current management relies on integrated strategies like vector control and removal of infected trees.
• Efficient detection of CG-infected trees is essential for effective disease management.

Purpose of the Study:

• To evaluate deep learning models for efficient Citrus Greening detection.
• To compare the performance of Faster R-CNN, VGGNet, and ResNet architectures.
• To assess the impact of the Convolution Block Attention Module (CBAM) on detection accuracy.

Main Methods:

• Developed Faster R-CNN models using transfer learning.
• Compared pre-trained VGGNet and ResNet models.
• Integrated Convolution Block Attention Module (CBAM) with VGGNet and ResNet to create VGGNet+CBAM and ResNet+CBAM variants.
• Deployed efficient models on web applications for farmer accessibility.

Main Results:

• ResNet-based models demonstrated superior performance in CG detection.
• The integration of CBAM significantly enhanced detection precision and overall model performance.
• Web applications enabled real-time, in-field diagnosis of CG disease.

Conclusions:

• Deep learning, particularly ResNet+CBAM, offers a highly efficient solution for Citrus Greening detection.
• The developed web applications provide a practical tool for farmers to diagnose CG in real-time.
• Improved detection facilitates timely intervention and better management of this destructive citrus disease.