The Gray Mold Spore Detection of Cucumber Based on Microscopic Image and Deep Learning

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Pathology

Background:

• Traditional pathogen spore detection is slow, labor-intensive, and subjective.
• Existing image processing methods struggle with complex scenes and manually designed features.
• Accurate spore detection is crucial for early disease diagnosis in precision agriculture.

Purpose of the Study:

• To develop a rapid and accurate detection algorithm for pathogen spores, specifically gray mold.
• To improve upon existing methods by addressing challenges in complex agricultural environments.
• To enhance the objectivity and efficiency of spore detection for disease management.

Main Methods:

• Proposed the MG-YOLO (Multi-head self-attention and Ghost-optimized YOLO) detection algorithm.
• Integrated Multi-head self-attention in the backbone for global information capture.
• Utilized weighted Bidirectional Feature Pyramid Network (BiFPN) for multi-scale feature fusion and GhostCSP for neck optimization.

Main Results:

• Achieved 0.983 accuracy in detecting gray mold spores, processing images in 0.009 seconds.
• Demonstrated effective detection of spores in challenging conditions: blurred, small targets, varied morphology, and high density.
• Improved detection accuracy by 6.8% compared to the YOLOv5 model.

Conclusions:

• The MG-YOLO algorithm offers a significant advancement in rapid and accurate pathogen spore detection.
• This method provides a novel approach to enhance the objectivity of spore detection in precision agriculture.
• The model meets the demand for high-precision spore detection, aiding in early disease diagnosis and management.