A classification method for soybean leaf diseases based on an improved ConvNeXt model.

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Deep learning models are increasingly used for plant disease detection.
• Current soybean leaf disease identification relies heavily on traditional machine learning methods.
• There is a need for more accurate and robust deep learning approaches for soybean disease diagnosis.

Purpose of the Study:

• To develop an enhanced deep learning network for accurate soybean leaf disease recognition.
• To improve the classification accuracy beyond existing deep learning and machine learning models.
• To validate the model's effectiveness on diverse plant leaf disease datasets.

Main Methods:

• Developed an enhanced deep learning network with feature extraction, attention calculation, and classification modules.
• Utilized data augmentation, including random masking, to improve dataset robustness.
• Incorporated an attention module with LeakyReLu activation for focused feature extraction and noise reduction.

Main Results:

• The enhanced network achieved an average soybean leaf disease recognition accuracy of 85.42%.
• This accuracy significantly surpassed six other deep learning models (ConvNeXt, ResNet50, Swin Transformer, MobileNetV3, ShuffleNetV2, SqueezeNet).
• The model also demonstrated strong performance on a grapevine leaf dataset, confirming its general applicability.

Conclusions:

• The proposed enhanced deep learning network effectively improves soybean leaf disease identification accuracy.
• The attention mechanism and data augmentation contribute to the model's superior performance.
• The model shows promise for broad application in automated plant disease detection systems.