Lightweight Sewer Pipe Crack Detection Method Based on Amphibious Robot and Improved YOLOv8n

Area of Science:

• Civil Engineering
• Computer Vision
• Robotics

Background:

• Underground urban sewage pipelines face challenges in crack detection due to difficult access and harsh conditions.
• Existing methods may lack efficiency and accuracy in identifying various crack types in complex environments.

Purpose of the Study:

• To develop a lightweight and efficient crack detection method for sewage pipelines.
• To enhance the performance of the YOLOv8n model for improved crack identification.
• To validate the proposed method in real-world sewage pipe scenarios.

Main Methods:

• Utilized sewage pipeline robots for rapid data collection in water and sludge media.
• Introduced a lightweight RGCSPELAN module to reduce model parameters.
• Replaced the detection head with Detect_LADH for improved feature extraction.
• Integrated the LSKA module into the SPPF module to enhance model robustness.

Main Results:

• The improved YOLOv8n model achieved 1.6 million parameters and an FPS of 261 for real-time detection.
• Demonstrated high detection accuracy for both small and long cracks.
• Outperformed other models like YOLOv5n, YOLOv6n, RT-DETRr18, YOLOv9t, and YOLOv10n in terms of parameter count and speed.

Conclusions:

• The proposed method is feasible for effective sewage pipe crack detection.
• The enhanced YOLOv8n model shows potential for improving safety, efficiency, and cost-effectiveness in urban sewage pipe maintenance.