ActiveGCN-CT: Active Learning-Enhanced Graph Convolutional Networks for Computed Tomography Inversion in Coal Mines

In the realm of coal mine safety, accurately detecting high-stress regions within the coal seam at the working face is paramount for averting rockburst incidents and boosting mining efficiency. This paper introduces ActiveGCN-CT, a novel graph convolutional network (GCN) model enriched with an active learning strategy, tailored for active source computational tomography (CT) inversion in coal mines. The research shows that the magnitude of stress in coal bodies is generally positively correlated with P-wave velocities; hence, precise P-wave velocity imaging can effectively pinpoint high-stress zones. In comparison to traditional CT inversion techniques and current deep learning tomography methods, ActiveGCN-CT delivers accurate P-wave velocity imaging even under conditions of sparse sensor networks and limited active seismic sources. By utilizing actual data from a coal mining face in China, our model employs an active learning strategy to efficiently select samples for training. This significantly improves the accuracy of seismic wave velocity model predictions and drastically reduces the number of training samples and iterations needed. Through comparisons with the SIRT algorithm and verification experiments conducted at coal mine sites, the effectiveness of the ActiveGCN-CT model in effectively identifying high-stress areas is confirmed, providing a new technical approach for the safety monitoring and management of coal mines.