Abstract: An intelligent identification and quantitative analysis method of rock fracture is proposed based on images,in order to realize the automated extraction and parametric characterization of fractures.Firstly,the D-LinkNet model framework was improved.The MD-LinkNet model is established by cascading cavity convolution (CAC) fusion with residual multicore pooling (RMP) module,which realized the effective extraction of multi-scale fracture feature information.Secondly,DUpsample bilinear interpolation function is utilized for up-sampling to optimize the low-resolution image of the fracture after convolutional pooling.The global and local features of the fracture are fused through the neighborhood pixel information to improve the accuracy of identification.Further,a two-dimensional fracture feature set is established based on the identification maps.A method of intersection identification and skeleton vectorization calculation for fractures is proposed,which extracts fracture parameters such as trace length,width,and apparent dip angle,and ultimately realizes the quantitative analysis of fracture identification images.Finally,a case study of tunnel fracture was carried out in Xinjiang,China.The results show that the accuracy of fracture identification is up to 95.7%,the precision is 85.1%.And the automatic extraction of geometrical parameters of tunnel rock mass is realized.This study provides guidance for the rapid intelligent identification and geometric parameter extraction of rock fracture in underground engineering.