Digital modeling optimization method of coal mine roadway based on attention mechanism and generative adversarial network

XUE Xusheng; QIN Yihan; YANG Xingyun; YUE Jianing; GUO Yifeng; MAO Qinghua; WANG Chuanwei; ZHANG Xuhui

doi:10.13247/j.cnki.jcumt.20240515

Abstract：Accurate modeling of underground roadway space environment in coal mine is an im-portant guarantee for the perception environment,autonomous positioning,navigation and control key technologies of coal mine mobile robots such as tunneling and inspection,and is al-so the key research and development direction of coal mine intelligent construction.At pres-ent,the accurate modeling of underground space environment in coal mines is faced with the problems of difficult feature perception,incomplete modeling information and low accuracy of model construction.In this paper,the modeling of coal mine roadway based on millimeter wave radar has some problems such as lack of point cloud and insufficient local reconstruction accuracy.A digital modeling optimization method of coal mine roadway based on attention mechanism and generative adversarial network was proposed.A"global+local"dual discrim-inator was constructed to improve the generative adversarial network model,optimize the dis-tribution and details of the generated network learning data,and provide accurate data for sol-ving the problem of insufficient point cloud reconstruction accuracy.In order to improve the a-bility of point cloud spatial geometric feature extraction,an improved discriminator method with spatial domain attention mechanism was proposed to accurately extract the point cloud ge-ometry and structural features required for roadway spatial modeling.The test results show that the average absolute error of the overall width of the reconstructed roadway after optimi-zation is 2.02 cm,which is 28％higher than that before optimization,and the maximum error is 7.40 cm,which is 14％higher than that before optimization.The average absolute error of the overall height of the roadway is 1.52 cm,which is 23％higher than that before optimiza-tion,and the maximum error is 4.10 cm,which is 13％higher than that before optimization.The proposed optimization method effectively improves the modeling accuracy of millimeter wave radar in complex coal mine roadway environment,which is of great value for intelligent,safe and efficient mining of coal mine robots based on high-precision coal mine roadway space environment.