Author Affiliations
College of Ocean and Space Information, China University of Petroleum (East China), Qingdao 266580, Shandong, Chinashow less
Fig. 1. Flow chart of LiDAR point cloud fracture zone extraction method
Fig. 2. Schematic diagram of OE convolution unit. (a) Point cloud in 3D space (the input point is at origin); (b) nearest neighbour search in eight octants; (c) convolution along X, Y, Z axis
Fig. 3. PointSIFT module
Fig. 4. Schematic diagram of three-dimensional convolution module framework
Fig. 5. Schematic diagram of PS-CNN point cloud fracture zone extraction framework
Fig. 6. Sample display diagrams of ISPRS point cloud datasets. (a); Samp51; (b) Samp53
Fig. 7. Sample display diagrams of Chuandian point cloud datasets. (a) CD_1; (b) CD_2
Fig. 8. Results of three fracture zone extraction methods on Samp51. (a) Label; (b) TD; (c) DNN; (d) PS-CNN
Fig. 9. Results of three fracture zone extraction methods on Samp53. (a) Label; (b) TD; (c) DNN; (d) PS-CNN
Fig. 10. Results of three fracture zone extraction methods on CD_1. (a) Label; (b) TD; (c) DNN; (d) PS-CNN
Fig. 11. Results of three fracture zone extraction methods on CD_2. (a) Label; (b) TD; (c) DNN; (d) PS-CNN
Error | Calculation method |
---|
T.I | c/(c+d) | T.II | b/(a+b) | T.E. | (b+c)/(a+b+c+d) |
|
Table 1. Confusion matrix of classification results and evaluation errors calculation method
Dataset | Method | T.Ⅰ/% | T.Ⅱ/% | T.E./% | Accuracy /% | Time /s |
---|
Samp51 | TD | 1.22 | 9.31 | 2.82 | 97.18 | 16 | DNN | 0.46 | 4.95 | 1.35 | 98.65 | 32 | PS-CNN | 0.40 | 2.43 | 0.79 | 99.21 | 51 | Samp53 | TD | 2.43 | 17.37 | 3.83 | 96.17 | 99 | DNN | 1.07 | 11.39 | 2.03 | 97.97 | 136 | PS-CNN | 0.18 | 10.16 | 1.11 | 98.89 | 193 |
|
Table 2. Performance comparison of three fracture zone extraction methods on ISPRS dataset
Dataset | Method | T.Ⅰ/% | T.Ⅱ/% | T.E./% | Accuracy /% | Time /s |
---|
CD_1 | TD | 1.15 | 1.47 | 1.32 | 98.68 | 238 | DNN | 0.77 | 0.97 | 0.88 | 99.12 | 342 | PS-CNN | 0.29 | 0.49 | 0.40 | 99.60 | 424 | CD_2 | TD | 1.39 | 1.31 | 1.36 | 98.64 | 273 | DNN | 0.63 | 2.01 | 1.28 | 98.72 | 364 | PS-CNN | 0.21 | 0.50 | 0.35 | 99.65 | 473 |
|
Table 3. Performance comparison of three fracture zone extraction methods on Chuandian dataset
Dataset | PointSIFT | T.Ⅰ /% | T.Ⅱ /% | T.E. /% | Aaccuracy /% |
---|
Samp51 | | 0.44 | 2.48 | 0.83 | 99.17 | √ | 0.40 | 2.43 | 0.79 | 99.21 | Samp53 | | 0.23 | 10.27 | 1.17 | 98.83 | √ | 0.18 | 10.16 | 1.11 | 98.89 |
|
Table 4. Performance comparison of three fracture zone extraction methods on ISPRS dataset
Dataset | Method | Dilution rate | Point | Aaccuracy /% | Time /s |
---|
CD_1 | PS-CNN | 0.3 | 65817 | 99.58 | 370 | 0.2 | 75363 | 99.60 | 424 | 0.1 | 84670 | 99.63 | 476 | CD_2 | PS-CNN | 0.3 | 72357 | 99.64 | 412 | 0.2 | 83059 | 99.65 | 473 | 0.1 | 93434 | 99.67 | 532 |
|
Table 5. Performance comparison of the proposed method on point cloud samples with different dilution rate