Miner pose detection algorithm based on improved YOLOv7 in the coal mine

Abstract

The unsafe behavior of workers in underground mines is a major cause of accidents. Recently, deep learning technology has become more and more widely used in the computer field and has achieved remarkable results, providing new opportunities for the research of miner behavior recognition algorithms. To address the current problem of large computation volume and low detection efficiency of multi-person posture detection in underground mines, we propose an improved algorithm that introduces the GSConv convolution module into the YOLOv7-Pose model and replaces the ELEN-W module with the GSELEN module to accelerate model convergence. We also introduce a group-convolution reconstruction SPPCSPC module to reduce parameter load and calculation amount and reduce the original 17 joint points to 14 joints to further realize a lightweight model. The experimental results showed that the improved YOLOv7-Pose model in this study decreased to 135.7 MB, by about 18.8%, while maintaining model accuracy. Therefore, this improvement can achieve a lightweight algorithm and improve real-time posture detection of construction workers in underground mines while ensuring model accuracy.