Colony-YOLO: A Lightweight Micro-Colony Detection Network Based on Improved YOLOv8n

Area of Science:

• Plant pathology
• Computer vision
• Agricultural technology

Background:

• Accurate detection of colony-forming units (CFUs) is crucial for mulberry bacterial blight research but is hampered by time constraints and detection inaccuracies.
• Existing methods for colony detection often struggle with small targets and high computational demands, limiting their practical application.

Purpose of the Study:

• To develop an efficient and accurate deep learning model for detecting mulberry bacterial blight colonies.
• To address the challenges of small-target detection and high computational consumption in automated colony counting.

Main Methods:

• A novel dataset, Mulberry Bacterial Blight Colony Dataset (MBCD), comprising 310 images with 23,524 colonies was created.
• A lightweight backbone network, StarNet, was integrated to reduce computational complexity.
• A modified C2f module (C2f-MLCA) incorporating Mixed Local Channel Attention (MLCA) was designed to enhance feature representation.
• The Shape-IoU loss function was employed to improve bounding box accuracy.

Main Results:

• The proposed Colony-YOLO model achieved a mean Average Precision (mAP) of 96.1% on the MBCD.
• Colony-YOLO demonstrated a 4.8% improvement in mAP compared to the baseline YOLOv8n.
• The model achieved reductions in computational load, with 1.8 G fewer FLOPs and 0.8 M fewer parameters.

Conclusions:

• Colony-YOLO effectively enhances detection accuracy for mulberry bacterial blight colonies while maintaining lower computational complexity.
• The developed model shows significant potential for practical applications in agricultural research and disease management.