Intelligent Detection Algorithm for Concrete Bridge Defects Based on SATH-YOLO Model.

Area of Science:

• Civil Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Traditional object detection models like YOLOv8 face challenges in bridge defect detection due to high computational demands and slow processing speeds.
• Intelligent construction and inspection necessitate more efficient and accurate automated defect identification systems.

Purpose of the Study:

• To propose a lightweight and efficient object detection model, SATH-YOLO, for bridge defect detection.
• To improve the performance of bridge inspection systems for resource-constrained environments and edge devices.

Main Methods:

• Developed the SATH-YOLO model by integrating the Star Block from StarNet into the STNC2f module for enhanced semantic enrichment and multi-scale feature fusion.
• Replaced the SPPF module with an AIFI module to capture finer-grained local features and improve feature fusion precision.
• Implemented a lightweight TDMDH detection head with shared convolution and dynamic feature selection to further reduce computational costs.

Main Results:

• Achieved significant reductions in parameter count (39.9%), computation (8.6%), and model size (36.2%) with the SATH-YOLO model.
• Maintained and improved average detection precision by 1% compared to traditional models.
• Demonstrated the model's suitability for edge devices and environments with limited computational resources.

Conclusions:

• The SATH-YOLO model offers a computationally efficient and accurate solution for bridge defect detection.
• The proposed model effectively balances performance and resource utilization, meeting the demands of intelligent inspection systems.