Target Detection Method for Soil-Dwelling Termite Damage Based on MCD-YOLOv8.

Area of Science:

• Engineering
• Environmental Science
• Computer Science

Background:

• Global climate change and ecological decline pose significant threats to hydraulic engineering safety.
• Soil-dwelling termites, like Odontotermes formosanus and Macrotermes barneyi, cause substantial damage to earth and rock structures such as reservoirs and embankments.
• Effective identification of termite damage is essential for proactive monitoring, early warning systems, and control strategies in hydraulic engineering.

Purpose of the Study:

• To develop an advanced object detection model for identifying soil-dwelling termite damage in hydraulic engineering.
• To enhance the accuracy and efficiency of termite damage detection using artificial intelligence.

Main Methods:

• An improved YOLOv8 model, MCD-YOLOv8, was developed incorporating a Monte Carlo attention (MCA) module and a dimension-aware selective integration (DASI) module.
• The MCA module enhances recognition of small targets by generating attention maps via random sampling pooling.
• The DASI module optimizes computation time and memory usage, improving detection speed and accuracy.

Main Results:

• The MCD-YOLOv8 model demonstrated superior performance compared to traditional and enhanced models in detecting termite damage.
• The model achieved significant improvements in precision (6.4%) and mean average precision (2.4%) over the standard YOLOv8.
• MCD-YOLOv8 reduced model parameters by 105,320, decreased detection box redundancy, and improved small target detection accuracy.

Conclusions:

• The MCD-YOLOv8 model offers a robust solution for intelligent identification of termite damage in complex hydraulic engineering environments.
• This AI-driven approach enhances intelligent monitoring of termite activity.
• The developed model provides crucial technical support for advancing termite control technologies in infrastructure protection.