Classification of Citrus Huanglongbing Degree Based on CBAM-MobileNetV2 and Transfer Learning.

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Citrus huanglongbing poses a significant challenge to fruit farmers in southern China, impacting agricultural development and farmer incomes.
• Accurate and timely diagnosis of citrus huanglongbing is crucial for effective disease management.

Purpose of the Study:

• To develop a novel classification model for diagnosing citrus huanglongbing using image recognition.
• To enhance the accuracy and efficiency of citrus huanglongbing detection through advanced deep learning techniques.

Main Methods:

• A CBAM-MobileNetV2 model was developed, integrating convolutional features with an attention module for enhanced information capture.
• Transfer learning, specifically parameter fine-tuning, was employed to optimize model performance.
• A dataset of 751 citrus huanglongbing images was augmented and split into training (80%) and testing (20%) sets.

Main Results:

• The CBAM-MobileNetV2 model achieved a high recognition accuracy of 98.75% for citrus huanglongbing images.
• Parameter fine-tuning in transfer learning outperformed parameter freezing, improving accuracy by 1.02-13.6%.
• The developed model demonstrated superior performance compared to standard MobileNetV2, Xception, and InceptionV3 models.

Conclusions:

• The CBAM-MobileNetV2 model combined with transfer learning provides a highly accurate image recognition solution for citrus huanglongbing.
• This approach offers a promising tool for early disease diagnosis, supporting sustainable citrus farming practices.