YOLOv11n-DualPC-Lite: a lightweight, high-precision real-time detection model for maize leaf diseases.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Maize leaf disease detection faces challenges in balancing model efficiency and accuracy, especially on resource-limited edge devices.
• Existing lightweight models often struggle to maintain high detection performance.

Purpose of the Study:

• To develop an enhanced lightweight target detection model, YOLOv11n-DualPC-Lite, for maize leaf disease identification.
• To improve model performance and reduce computational load for edge device deployment.

Main Methods:

• Designed the C2fDualPConv module with PartialConv to enhance feature representation and reduce parameters.
• Introduced a Slim-Neck architecture with VoVGSCSPC_SimAm for parameter reduction and feature strengthening.
• Implemented an EfficientHead detection head using an inverted bottleneck MBConv module for efficient feature extraction.

Main Results:

• The YOLOv11n-DualPC-Lite model achieved a mAP50 score of 90.9%, outperforming other lightweight models.
• The model has 2.13 million parameters, 4.55 G computational complexity, and a size of 4.41 MB.
• Achieved a 1.9% higher mAP50 than YOLOv11n, with a 17.8% reduction in parameters and 29.3% less computational complexity.

Conclusions:

• YOLOv11n-DualPC-Lite offers a superior balance between detection accuracy and lightweight performance for maize leaf disease monitoring.
• The model provides an efficient and practical solution for real-time crop disease detection on edge devices.
• Demonstrated significant improvements in detection speed (27.8% increase) on a Raspberry Pi 5.