Intelligent identification and extraction of geometric parameters for surface fracture networks of rocky slopes

As an important prerequisite for modeling the high steep rocky slope, a fast and accurate parametric modeling for fracture networks of rocky slopes has become a popular research topic in recent years. Focusing on the deep learning and intelligent algorithmic clustering method, a UAV photography-based joint detection technique is proposed to identify and extract the geometric parameters of the fracture network on high steep slope surface. A dilated convolution is adopted to improve the traditional U-net segmentation network, and a GMM-EM algorithm is employed to cluster the segmented fractures on the binary images. Finally, a RANSAC algorithm is used to perform the extraction process of geometric parameter of the fracture network. Seen from the comparative results of DICE similar index, the accuracy of segmentation recognition is more than 97%, which shows that the proposed fracture extraction technique is more efficient and accurate than other traditional algorithms. The improved technique is applied to the slope of Lukuishan open pit, implementing the in-site rapid data extraction of fracture networks on the slope surface. This technique may provide an effective technical support for the refined modeling of high and steep rocky slopes.