ConvGeM-next: a deep learning framework for plant disease detection.

Area of Science:

• Agricultural Science
• Computer Science
• Machine Learning

Background:

• Plant diseases significantly threaten sustainable agriculture and food security.
• Accurate and early disease identification is vital for minimizing crop losses.
• Existing deep learning models, like CNNs, face generalization challenges in real-world conditions due to visual variations and noise.

Purpose of the Study:

• To develop an advanced deep learning architecture for fine-grained plant disease classification with enhanced generalization.
• To improve the accuracy and reliability of automated plant disease diagnosis systems.

Main Methods:

• Introduced ConvGem-NeXt, an end-to-end deep learning model based on the ConvNeXt architecture.
• Incorporated a learnable Generalized Mean pooling layer and ReLU activation for better spatial feature representation.
• Utilized a custom classifier head with batch normalization, ReLU, and dropout to prevent overfitting and boost accuracy.

Main Results:

• Achieved 99.65% accuracy on the PlantVillage dataset.
• Attained 94.69% accuracy on the challenging, real-world PlantDoc dataset.
• Demonstrated the model's efficacy in reliably classifying plant diseases under varied conditions.

Conclusions:

• ConvGem-NeXt offers a robust framework for timely plant disease diagnosis.
• The model contributes to agricultural automation by enhancing crop productivity through reliable disease identification.