ADQ-YOLOv8m: a precise detection model of sugarcane disease in complex environment

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Current sugarcane disease identification methods are limited in scope and applicability.
• A need exists for more precise and versatile disease detection systems in sugarcane cultivation.

Purpose of the Study:

• To develop an enhanced object detection model, ADQ-YOLOv8m, for accurate sugarcane disease identification.
• To improve feature representation and address class imbalance in sugarcane disease detection.

Main Methods:

• Modified YOLOv8m framework with a Dynamic Head for enhanced feature representation.
• Incorporated ATSS dynamic label assignment and QFocalLoss to handle class imbalance.
• Utilized visual analysis and cross-scenario adaptability testing for comprehensive evaluation.

Main Results:

• ADQ-YOLOv8m achieved superior performance compared to nine other object detection models.
• Key performance metrics include 86.90% precision, 85.40% recall, 90.00% mAP50, and 86.00% F1 score.
• The model demonstrated excellent multi-objective processing and strong generalization capabilities.

Conclusions:

• The ADQ-YOLOv8m model offers robust and accurate sugarcane disease detection, even in complex scenarios.
• It provides strong support for intelligent sugarcane cultivation and effective disease management strategies.
• The model's generalization capability makes it suitable for diverse agricultural applications with class imbalance.