Crack identification and detection in mine boreholes using deep learning: A case study of hexi coal mine

The internal cracks in deep rock masses are an important factor determining the safety of coal mining. The borehole peeping technology is an important method for obtaining the distribution of coal and rock fractures in deep geological engineering. However, there are often problems such as strong background interference and complex joints in real mines, which bring greater challenges to obtaining the distribution of cracks. Therefore, this study proposes a deep learning-based method for identifying internal cracks, which is used for the intelligent identification and analysis of cracks in drilled rock masses. The experimental verification was carried out based on the real borehole peeping images of the tunnels in Hexi Coal Mine as the data set. By embedding the ResCBAM attention mechanism into the YOLOv8 head network, the reliability and accuracy of identifying internal cracks in rock were further enhanced. The test results show that, the proposed Res-YOLOv8 algorithm can more accurately suppress the interference of complex backgrounds and present the morphological characteristics of rock cracks to compare with the fusion of typical attention mechanisms SE and CBAM with YOLOv8. The ResCBAM-YOLOv8 algorithm demonstrates strong performance in the evaluation indicators of confusion matrix, loss rate, precision, recall and MPA. The proposal of method has improved the efficiency of detecting and identifying cracks in coal and rock masses in real mines, and provided more reliable technical support for the stability evaluation of coal mine roof.