A Deep Learning‐Based Method for Recognition and Classification of Fracture Images in Coal Mine Surrounding Rock

ABSTRACT

Accurately identifying fractures in surrounding rock is crucial for ensuring the stability of coal mine roadways and safe mining operations. Traditional methods lack sufficient accuracy and adaptability, while existing deep learning approaches face a trade‐off between high computational cost and inadequate feature extraction, and the scarcity of relevant public datasets has hindered research development. To address this, this paper proposes a deep learning‐based method for the image recognition and classification of coal roadway surrounding rock fractures, which balances high performance with low computational overhead. A lightweight plus convolutional block (LPCB) integrating depthwise separable convolution and asymmetric expanded convolution is designed to reduce computational cost while enlarging the receptive field and enhancing fracture feature capture capabilities. To address the issue of limited datasets, we independently constructed a labeled dataset containing images of surrounding rock fractures from different mines and varying geological conditions. Based on this module, a recognition framework was developed and experimentally validated, and the detection targets were innovatively classified to provide quantitative evidence for the causes and evolution mechanisms of the fractures. The results demonstrate that the proposed framework achieves a good balance between accuracy and efficiency, with a precision of 90.3%, recall of 89.1%, mAP50 of 92.4%, and a frame rate of 23.7 FPS. This study provides a reliable technical solution for rapid and precise detection of surrounding rock fractures and holds significant theoretical and practical value for the intelligent development of underground engineering safety monitoring.