DSA-net: a lightweight and efficient deep learning-based model for pea leaf disease identification.

Area of Science:

• Agricultural Science
• Computer Vision
• Deep Learning

Background:

• Pea leaf diseases significantly reduce crop yield and quality.
• Current identification methods lack efficiency, real-time processing, and cost-effectiveness for modern agriculture.

Purpose of the Study:

• To develop a deep learning model for accurate and efficient pea leaf disease identification.
• To address limitations of existing methods in feature extraction, environmental sensitivity, and scalability.

Main Methods:

• Proposed DSA-Net model integrating an improved MobileNet-V3_small architecture.
• Incorporated deformable convolution for geometric feature modeling.
• Integrated self-attention and additive attention mechanisms for enhanced feature recognition.

Main Results:

• Achieved an average recognition accuracy of 99.12% on a dataset of 7915 pea leaf samples.
• The model has a compact parameter size of 1.48M, suitable for efficient deployment.
• Successfully classified five categories: healthy, brown spot, leaf miner, powdery mildew, and root rot.

Conclusions:

• The DSA-Net model significantly enhances pea leaf disease diagnostic accuracy.
• The approach shows potential for edge device deployment in agricultural settings.
• Offers a scalable, accurate, and cost-effective solution for disease management in agriculture.