EADD-YOLO: An efficient and accurate disease detector for apple leaf using improved lightweight YOLOv5

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Existing apple leaf disease detection methods face challenges with high parameter counts, slow speeds, and poor performance on small, dense spots.
• These limitations hinder the practical application of automated disease detection in agriculture.

Purpose of the Study:

• To develop an efficient and accurate model for apple leaf disease detection.
• To address the limitations of current methods by optimizing for speed, accuracy, and parameter efficiency.

Main Methods:

• Proposed EADD-YOLO model based on YOLOv5, incorporating a lightweight shufflenet inverted residual module in the backbone.
• Introduced a depthwise convolution-based feature learning module in the neck network for efficient feature extraction and fusion.
• Integrated a coordinate attention module to focus on critical disease spot information and suppress irrelevant data.
• Utilized SIoU loss for improved bounding box regression accuracy.

Main Results:

• Achieved 95.5% mean average precision (mAP) and 625 frames per second (FPS) on the apple leaf disease dataset (ALDD).
• Demonstrated significant improvements over the latest research methods on ALDD, with a 12.3% increase in accuracy and 596 FPS.
• Showcased substantially lower parameter quantity and FLOPs compared to other popular algorithms.

Conclusions:

• The EADD-YOLO model offers a high-performance solution for early apple leaf disease diagnosis.
• The model's efficiency and accuracy make it suitable for deployment in agricultural robots.
• The open-sourced code facilitates further research and application in precision agriculture.