Concrete slump detection based on light-AWGAM-YOLOv8n.

Area of Science:

• Construction Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Precise concrete mortar slump detection is crucial for intelligent concrete mixing, impacting project quality and construction efficiency.
• Existing methods may lack the required efficiency and precision for real-time applications.
• The YOLOv8n algorithm offers a foundation for object detection but requires optimization for specific tasks.

Purpose of the Study:

• To improve the efficiency and precision of concrete mortar slump detection.
• To enhance the YOLOv8n algorithm for better feature fusion and reduced computational load.
• To balance detection accuracy with computational efficiency in intelligent construction.

Main Methods:

• Integrated Add Weight Global Attention Mechanism (AWGAM) with C2f modules (Basic and Adaptive versions) in the YOLOv8n neck and backbone networks.
• Introduced depth-separable convolution into AWGAM to create a lightweight module (Light-AWGAM) for the backbone.
• Deployed enhanced modules to improve multi-scale feature fusion and model efficiency.

Main Results:

• The improved YOLOv8n model demonstrated a 3.3% increase in precision and a 0.6% increase in recall.
• Mean Average Precision (mAP50 and mAP50-95) showed improvements of 0.8%.
• The model achieved a significant reduction in parameters while maintaining high detection accuracy and computational efficiency.

Conclusions:

• The enhanced YOLOv8n algorithm with AWGAM significantly improves concrete mortar slump detection.
• The proposed lightweight module and attention mechanisms effectively balance accuracy and efficiency.
• This optimized model is well-suited for efficient and precise detection tasks in intelligent concrete mixing.