Wheat disease recognition method based on the SC-ConvNeXt network model.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Wheat disease identification using convolutional neural networks (CNNs) requires extensive labeled data, which is costly and time-consuming.
• Natural environmental factors complicate accurate disease recognition in wheat.
• Existing models often struggle with data scarcity and environmental variability.

Purpose of the Study:

• To develop an efficient wheat disease identification model that minimizes reliance on labeled data.
• To improve the robustness and accuracy of wheat disease recognition in complex natural settings.
• To enhance feature extraction and generalization capabilities for agricultural image analysis.

Main Methods:

• Proposed SC-ConvNeXt model integrating SimCLR self-supervised pre-training with ConvNeXt-T.
• Incorporated an improved CBAM attention mechanism with LeakyReLU activation in the loss function.
• Utilized Focal Loss to handle class imbalance and employed data augmentation techniques.
• Evaluated model performance against four classic classification models using a wheat disease dataset.

Main Results:

• The SC-ConvNeXt model achieved the highest average classification accuracy of 88.05% on the test set.
• The integration of SimCLR significantly reduced the need for labeled training data.
• The attention mechanism enhanced feature extraction in complex backgrounds, improving generalization.
• The model demonstrated superior performance compared to traditional classification models.

Conclusions:

• The SC-ConvNeXt model effectively identifies wheat diseases with high accuracy, even with limited labeled data.
• Self-supervised learning and attention mechanisms are crucial for robust agricultural disease detection in real-world conditions.
• The proposed approach offers a cost-effective and efficient solution for automated wheat disease monitoring.