Vegetable disease detection using an improved YOLOv8 algorithm in the greenhouse plant environment

Area of Science:

• Agricultural Technology
• Computer Vision
• Plant Pathology

Background:

• Accurate detection of vegetable diseases in greenhouses is crucial for crop management.
• Existing object detection models face challenges with imprecise detection, small objects, and varying scales.

Purpose of the Study:

• To develop an optimized object detection model (YOLOv8n-vegetable) for enhanced vegetable disease identification in greenhouse environments.
• To improve detection accuracy, speed, and efficiency for real-time disease monitoring.

Main Methods:

• Integration of a novel C2fGhost module using GhostConv for reduced parameters and improved performance.
• Incorporation of the Occlusion Perception Attention Module (OAM) to preserve feature information during fusion.
• Addition of a specific layer for small object detection and implementation of HIoU boundary loss for better convergence and regression.

Main Results:

• The enhanced YOLOv8n-vegetable model achieved a 6.46% increase in mean average precision (mAP) on a self-built greenhouse vegetable disease dataset.
• Model parameters and size were reduced by 0.16G and 0.21 MB, respectively.
• Demonstrated improved detection speed, accuracy, and real-time capability compared to the original model and other advanced detectors.

Conclusions:

• YOLOv8n-vegetable offers significant advancements in vegetable disease detection accuracy and efficiency.
• The lightweight and fast nature of the model makes it highly competitive and suitable for practical greenhouse applications.
• This model provides a promising solution for automated and real-time monitoring of vegetable diseases.