MGA-YOLO: A lightweight one-stage network for apple leaf disease detection

Area of Science:

• Agricultural Science
• Computer Vision
• Artificial Intelligence

Background:

• Apple leaf diseases significantly impact crop yield and quality.
• Current diagnostic methods lack efficiency and accuracy.
• Existing computer vision models require high-performance hardware, limiting field application.

Purpose of the Study:

• To develop a lightweight, real-time apple leaf disease detection model for mobile devices.
• To improve the accuracy and efficiency of on-site disease diagnosis.
• To address the limitations of current computer vision approaches in agricultural settings.

Main Methods:

• Proposed a lightweight one-stage network, Mobile Ghost Attention YOLO (MGA-YOLO).
• Utilized Ghost modules and Mobile Inverted Residual Bottleneck Convolution for efficiency.
• Integrated Convolutional Block Attention Module (CBAM) and an extra prediction head for enhanced feature extraction and object detection.
• Built the Apple Leaf Disease Object Detection (ALDOD) dataset with 8,838 images.

Main Results:

• MGA-YOLO achieved a 89.3% mAP on the ALDOD testing set, outperforming state-of-the-art methods.
• The model boasts a small size (10.34 MB) and high FPS (84.1 on GPU, 12.5 on mobile).
• Image augmentation further improved mAP to 94.0%.

Conclusions:

• MGA-YOLO offers an accurate and efficient solution for real-time apple leaf disease detection on mobile devices.
• The model's lightweight design makes it suitable for field deployment.
• This technology has the potential to significantly aid farmers in timely disease management and improve crop productivity.