Abstract In this paper, the fracture identification and feature extraction of tunnel face images are studied. Firstly, the data enhancement including various illumination processing measures is carried out on the tunnel face image set based on the characteristics of insufficient illumination and uneven light in the tunnel. Though recognizing the contour of the tunnel face by the Unet network, the average value of intersection over union and the average similarity are 91% and 93%. The morphological operation is used to make the edge of the tunnel face contour smooth and eliminate noise points. Then, the high-resolution tunnel face images are processed by splitting-splicing strategy, and the rock mass fractures are identified by DeepCrack network model transfer learning. The average value of intersection over union and average similarity are 61% and 75%. Further, the identification results are refined, skeletonized and analyzed in connected domains by using the Zhang-Suen algorithm and the 8 neighborhoods labeling algorithm. Finally, the pixel-level length and dip angle of each fracture is calculated by the control point marking and corrosion marking methods.
Abstract：
In this paper, the fracture identification and feature extraction of tunnel face images are studied. Firstly, the data enhancement including various illumination processing measures is carried out on the tunnel face image set based on the characteristics of insufficient illumination and uneven light in the tunnel. Though recognizing the contour of the tunnel face by the Unet network, the average value of intersection over union and the average similarity are 91% and 93%. The morphological operation is used to make the edge of the tunnel face contour smooth and eliminate noise points. Then, the high-resolution tunnel face images are processed by splitting-splicing strategy, and the rock mass fractures are identified by DeepCrack network model transfer learning. The average value of intersection over union and average similarity are 61% and 75%. Further, the identification results are refined, skeletonized and analyzed in connected domains by using the Zhang-Suen algorithm and the 8 neighborhoods labeling algorithm. Finally, the pixel-level length and dip angle of each fracture is calculated by the control point marking and corrosion marking methods.
LUO Hu1 Miller Mark1 ZHANG Rui2 FANG Yong1
.Research on the Automatic Identification Method for Rock Mass Fracture in Tunnel Face Based on Computer Vision Technology and Deep Learning[J] MODERN TUNNELLING TECHNOLOGY, 2023,V60(1): 56-65
2023, Vol. 60 