Cotton leaf disease detection model focusing on small targets and comprehensive feature extraction

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Pathology

Background:

• Accurate cotton leaf disease detection is vital for crop yield and sustainable agriculture.
• Small lesion sizes pose challenges for traditional detection methods, leading to missed or false detections.

Purpose of the Study:

• To develop an improved YOLOv8-based model (CM-YOLO) for enhanced detection of small cotton leaf disease targets.
• To increase the accuracy and robustness of automated disease detection systems in agriculture.

Main Methods:

• Integration of the SS2D module from VMamba into the backbone for multi-directional feature extraction.
• Incorporation of the MSDA module to optimize focus on small targets and reduce redundant computations.
• Replacement of the original bounding box loss with DIoU loss for precise localization and faster convergence.

Main Results:

• CM-YOLO achieved a mean Average Precision at 50% IoU (mAP50) of 0.933 and a recall of 0.891.
• Outperformed state-of-the-art models YOLOv8n (mAP50: 0.874) and YOLOv11n (mAP50: 0.930).
• Demonstrated high detection accuracy and robustness across diverse plant datasets in generalization experiments.

Conclusions:

• The proposed CM-YOLO model significantly enhances the detection performance for small cotton leaf disease targets.
• The model's effectiveness and applicability in complex agricultural scenarios are validated.
• CM-YOLO offers a valuable reference for intelligent agricultural disease detection research.