LCAMNet: a lightweight model for apple leaf disease classification in natural environments

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Apple leaf diseases significantly impact crop yield and quality.
• Accurate disease classification is vital for effective management.
• Existing models struggle with lightweight design and high accuracy in natural environments due to background-lesion similarity.

Purpose of the Study:

• To develop a lightweight and accurate model for apple leaf disease classification.
• To address the limitations of existing models in natural orchard settings.
• To improve the practical applicability of disease detection systems.

Main Methods:

• Introduction of the lightweight converged attention multi-branch network (LCAMNet).
• Integration of depthwise separable convolutions and structural re-parameterization for efficient modeling.
• Design of a dual-branch downsampling module to prevent feature loss.
• Implementation of a multi-scale structure and improved triplet attention for enhanced feature representation.

Main Results:

• LCAMNet achieved 92.60% accuracy on the constructed SCEBD dataset and 95.31% on a public dataset.
• The model demonstrates a lightweight design with only 0.03 GFLOPs and 1.30M parameters.
• The network effectively handles complex natural environments with various interference factors.

Conclusions:

• LCAMNet offers a high-accuracy, lightweight solution for apple leaf disease classification.
• The model's efficiency and accuracy make it suitable for deployment on resource-constrained devices in real-world orchards.
• The developed SCEBD dataset realistically represents orchard conditions, aiding future research.