Tea leaf disease and insect identification based on improved MobileNetV3

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Accurate identification of tea leaf diseases and insects is vital for maintaining tea quality and yield.
• Traditional identification methods are labor-intensive and not scalable for widespread application.

Purpose of the Study:

• To develop an accurate and efficient automated system for recognizing tea leaf diseases and insects.
• To enhance the MobileNetV3 architecture for improved performance in this specific task.

Main Methods:

• A dataset of 17 tea leaf disease and insect types was created, incorporating data augmentation.
• MobileNetV3 was improved by integrating the Coordinate Attention (CA) module.
• Transfer learning and fine-tuning strategies were applied to optimize the model.

Main Results:

• The improved MobileNetV3-CA model with transfer learning achieved a recognition accuracy of 95.88%.
• Comparative analysis showed superior performance against original MobileNetV3 and other classical models (ResNet18, AlexNet, VGG16, SqueezeNet, ShuffleNetV2).
• The developed application demonstrated robustness for intelligent diagnosis.

Conclusions:

• The enhanced MobileNetV3-CA model offers a highly accurate and robust solution for tea leaf disease and insect identification.
• This automated approach provides a reliable reference for intelligent diagnosis in tea cultivation.
• The study highlights the potential of deep learning for improving agricultural pest and disease management.