ALD-YOLO: a lightweight attention detection model for apple leaf diseases

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Apple cultivation is vital economically, but susceptible to diseases causing significant yield and financial losses.
• Accurate and rapid detection of apple leaf diseases is crucial for effective management and mitigation strategies.
• Existing detection methods may lack the efficiency or accuracy required for large-scale agricultural applications.

Purpose of the Study:

• To develop a lightweight and efficient deep learning model for the rapid and accurate detection of apple leaf diseases.
• To optimize the YOLOv8 architecture for improved performance in terms of accuracy and computational efficiency.
• To enhance the model's capability in identifying multi-scale features and small disease targets.

Main Methods:

• Proposed ALD-YOLO, a lightweight attention detection model based on the YOLOv8 architecture.
• Introduced Faster_C2F module by optimizing YOLOv8's C2F modules with FasterNet Block for improved efficiency.
• Integrated the EMA (Efficient Multi-Scale Attention) module to enhance multi-scale feature capture and focus on smaller disease targets, creating the Faster_C2F_EMA module.
• Replaced convolutional modules with ADown, a lightweight downsampling module, to reduce parameters and retain image information.

Main Results:

• ALD-YOLO demonstrated increased mAP (mean Average Precision) by 1.4% over YOLOv8n and 0.6% over YOLOv8s on the AppleLeaf9 dataset.
• Achieved significant reductions in GFLOPs (Giga Floating-point Operations Per Second): 29.63% compared to YOLOv8n and 79.93% compared to YOLOv8s.
• CPU inference testing showed up to a 119.23% improvement in frames per second compared to YOLOv8s.
• The model achieved an optimal balance between detection accuracy and computational efficiency.

Conclusions:

• The ALD-YOLO model offers a stable and efficient solution for detecting apple leaf diseases.
• The proposed model's lightweight nature and improved performance make it suitable for deployment on edge devices.
• ALD-YOLO contributes to advancing precision agriculture through enhanced disease detection capabilities.