Deep learning for mango leaf disease identification: A vision transformer perspective.

Area of Science:

• Agricultural Science
• Computer Science
• Machine Learning

Background:

• Machine learning, especially deep learning with Convolutional Neural Networks (CNNs), is increasingly used in smart agriculture for early plant disease detection.
• Vision Transformers (ViTs) have emerged as powerful tools for image classification, often surpassing traditional CNN performance.
• The application of ViTs in agriculture is a developing field.

Purpose of the Study:

• To evaluate the performance of Vision Transformers (ViTs) for identifying mango leaf diseases.
• To compare ViT performance against established CNN models in this agricultural application.
• To introduce an optimized ViT model for enhanced disease identification accuracy and efficiency.

Main Methods:

• Utilized a pretrained Data-efficient Image Transformer (DeiT) architecture, optimized for mango leaf disease identification.
• Compared the performance of the optimized ViT model against several popular CNN architectures (SqueezeNet, ShuffleNet, EfficientNet, DenseNet121, MobileNet).
• Developed a mobile application integrating the ViT model for real-time disease diagnosis.

Main Results:

• The optimized ViT model achieved a high accuracy of 99.75% in identifying mango leaf diseases.
• ViTs demonstrated superior performance compared to the evaluated CNN models.
• Vision Transformers exhibited shorter training times, requiring fewer epochs to reach optimal results than CNNs.

Conclusions:

• Vision Transformers represent a highly effective and efficient approach for mango leaf disease identification in smart agriculture.
• The developed ViT model and mobile application offer a promising solution for real-time disease management.
• Further adoption of ViTs in agriculture can significantly enhance crop monitoring and disease detection capabilities.