An efficient convolutional neural network-based diagnosis system for citrus fruit diseases.

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Fruit diseases significantly impact agricultural economic returns.
• Deep learning models are increasingly utilized for crop disease identification and severity diagnosis.
• High-latitude feature extraction is crucial for improving classification performance in agriculture.

Purpose of the Study:

• To leverage deep convolutional neural networks for enhanced citrus fruit disease identification and severity diagnosis.
• To improve multi-scale feature extraction and classification accuracy using a novel deep learning architecture.
• To develop an effective method for diagnosing the severity of citrus fruit diseases.

Main Methods:

• A hybrid deep learning model combining Inception module and EfficientNetV2 for feature extraction.
• Utilizing VGG to replace the U-Net backbone for improved segmentation performance.
• Applying transfer learning to optimize network training for disease detection and severity diagnosis.

Main Results:

• The proposed method achieved over 95% accuracy in citrus fruit disease recognition.
• The VGG-U-Net model demonstrated superior segmentation performance with 87.66% accuracy.
• The integrated approach proved effective for both disease identification and severity level diagnosis.

Conclusions:

• The developed deep learning method is highly effective for identifying and diagnosing citrus fruit diseases and their severity.
• The combination of EfficientNetV2 and Inception modules enhances multi-scale feature extraction.
• VGG-U-Net shows promising results for precise disease segmentation in agricultural applications.