Detection and identification of centipedes based on deep learning.

Area of Science:

• Agricultural Technology
• Computer Vision
• Wildlife Monitoring

Background:

• Accurate centipede population quantification is crucial for intelligent management and industry upgrades.
• Existing centipede counting methods suffer from low accuracy, large model sizes, and poor mobile deployment, hindering practical use.

Purpose of the Study:

• To develop a lightweight and efficient centipede detection model (FCM-YOLO) for improved accuracy and mobile applicability.
• To address limitations of current models in speed, size, and deployment.

Main Methods:

• Utilized the YOLOv5s framework, incorporating a C3FS module for reduced parameters and increased speed.
• Integrated a CBAM attention module to enhance target focus and suppress irrelevant information.
• Proposed a novel CMPDIOU loss function for precise bounding box localization.

Main Results:

• FCM-YOLO achieved a detection accuracy of 97.4%, a 2.7% improvement over YOLOv5s.
• Reduced floating-point operations (FLOPs) to 11.5G, a decrease of 4.3G compared to YOLOv5s.
• Demonstrated reduced parameter size and increased processing speed.

Conclusions:

• FCM-YOLO offers a significant advancement in centipede detection and enumeration.
• The model's lightweight design and high accuracy support practical deployment in intelligent agricultural systems.
• Contributes to the modernization of the rural centipede industry through enhanced technological capabilities.