Working face status detection in coal mine based on YOLOv8-EST

Area of Science:

• Computer Science
• Artificial Intelligence
• Deep Learning

Background:

• Deep learning object detection is expanding in coal mines.
• Increasing accuracy demands challenge limited computational resources in mining environments.
• Existing methods often increase network depth and parameters, exacerbating resource limitations.

Purpose of the Study:

• To propose a novel object detection algorithm, YOLOv8-EST, optimized for coal mining faces.
• To enhance feature extraction and spatial relationship modeling while reducing computational complexity.
• To improve detection accuracy and efficiency for real-time applications in coal mining.

Main Methods:

• Introduced Swin Transformer module into YOLOv8 for enhanced feature extraction.
• Modified GELU activation function for reduced computational complexity and improved training performance.
• Implemented relative position encoding using a deep learning network for better spatial relationship capture.
• Integrated EMA Attention mechanism to focus on critical image regions and relevant features.

Main Results:

• YOLOv8-EST demonstrated significant improvements in detection accuracy and efficiency on a custom coal mining dataset.
• The algorithm achieved high accuracy while reducing computational requirements and model size.
• Comparative analysis confirmed the algorithm's efficiency and usefulness in coal mining scenarios.

Conclusions:

• YOLOv8-EST offers an ideal solution for real-time, efficient, and accurate object detection in coal mining faces.
• The algorithm effectively balances high detection performance with reduced computational load.
• This contributes to enhanced production safety and efficiency in the coal mining industry.