This invention discloses a method for identifying fractures in roadway surrounding rock based on semantic segmentation and a joint
loss function. The method includes constructing and labeling three types of labeled borehole fracture datasets; data augmentation and pixel binarization
equalization preprocessing; pixel-level segmentation of the augmented borehole dataset; introducing a two-stage transfer learning training model; using a trained and improved Deeplabv3+ model to identify borehole
fracture test sets, and evaluating the segmentation effect using MIoU and PA; batch importing the predicted fracture segmentation images of different types into ImageJ
software to set pixel legend scales and obtain
actual length parameters. This method reduces the model size while ensuring accurate and rapid extraction of depth feature information of annular fractures, longitudinal fractures, and fractured zones from the rock strata, and obtaining microscopic fracture quantification indicators. It achieves automated identification of borehole fractures in surrounding rock, providing a more reliable reference for evaluating roadway surrounding rock
instability.