Vision-Based Detection of Large Coal Fragments in Fully Mechanized Mining Faces Using Adaptive Weighted Attention and...

Area of Science:

• Mining Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Scraper conveyor unloading ports are vital in mechanized mining but susceptible to blockages from large coal fragments.
• Existing visual perception methods for crushing robots lack accuracy and real-time performance in complex mining environments.
• Accurate identification of large coal fragments is crucial for preventing operational disruptions.

Purpose of the Study:

• To develop an advanced visual inspection method for coal mine crushing robots to accurately detect large coal fragments.
• To enhance the real-time performance and accuracy of visual perception systems in mining operations.
• To address the limitations of current methods in identifying large coal pieces in challenging environments.

Main Methods:

• Proposed LCDet method utilizing transfer learning and an adaptive weighted attention mechanism.
• Designed a lightweight backbone network with grouped convolution for efficient feature representation.
• Implemented an adaptive weighted attention mechanism to focus on coal fragments and reduce background noise.

Main Results:

• LCDet achieved high accuracy and recall on public and self-constructed datasets, outperforming YOLOv8n.
• Demonstrated a favorable balance between detection accuracy and model complexity, suitable for robotic deployment.
• Experimental results showed LCDet's effectiveness in detecting large coal fragments with superior performance and low parameter count.

Conclusions:

• LCDet enables lightweight and accurate detection of large coal fragments, crucial for automated mining operations.
• The method supports real-time deployment on crushing robots, enhancing efficiency and safety in mechanized mining.
• LCDet offers a promising solution for preventing blockages and improving overall mining productivity.