Research and Application of an Improved YOLOv7-based Personnel Intrusion Detection Method for Dangerous Areas in Underground Mines

To address the issue of underground workers entering the yellow-line restricted area (dangerous area) and meet safety standard requirements, we propose a personnel intrusion detection method based on an improved YOLOv7 object detection model. The improvements to the YOLOv7s model include: first, integrating attention mechanisms into the model's backbone layer to enhance its ability to extract features of small targets; second, improving the regression loss function to enhance the accuracy of bounding box localization. These improvements effectively address the identification rate of bounding boxes in densely populated work environments and improve the accuracy of detecting region intrusions. Additionally, we constructed an image sample library specifically for the underground mine working area, conducted sample cleaning to ensure the informational richness of the sample set. Experimental analysis shows that the improved YOLOv7 model effectively detects personnel intrusion into restricted areas, with an average precision improvement of 4.31% compared to the original YOLOv7 model, effectively reducing false positives and false negatives. Moreover, the PNPloy algorithm is used to implement the detection of dangerous area intrusion. This method has been successfully applied and practiced in engineering.