A point cloud segmentation method for power lines and towers based on a combination of multiscale density features and point-based deep learning

Figures & data

Figure 1. Point data: (a1–f1) Top views of the point clouds in sections 1–6. (a2–f2) The side view of point clouds in sections 1–6. (g1) Different types of power towers. (g2) Telegraph poles. (h) Different types of power lines (cross-section of bundled lines in the red box).

Table 1. Data details.

No.	Number of point clouds	Area (m^2)	Average density (pts/m^2)	Capacity (GB)	Remarks
1	4.85 × 10^7	2.89 × 10^5	167.91	1.17	Large elevation difference, sparse objects
2	3.22 × 10^6	1.94 × 10^5	16.58	0.08	Large elevation difference, sparse objects, tall buildings
3	4.50 × 10^7	2.04 × 10^5	220.85	1.23	Small elevation difference, dense objects
4	1.40 × 10^7	6.75 × 10^5	20.73	0.45	Small elevation difference, dense objects, less vegetation
5	4.51 × 10^7	3.21 × 10^5	140.89	0.86	Large elevation difference, sparse objects
6	2.57 × 10^7	4.22 × 10^5	60.84	0.69	Small elevation difference, dense objects, tall buildings and vegetation

Figure 2. Different densities of point clouds: The average densities of (a–d) are 45.8, 21.5, 9.4 and 4.5 pts/m², respectively. The average densities of (e–h) are 30.6, 13.1, 5.8 and 1.9 pts/m², respectively (the red lines show the basic shape and structure of the objects).

Figure 3. The density characteristics of point clouds (rotating a block of the point clouds along the direction of the structural points).

Figure 4. The flowchart of the segmentation method.

Figure 5. The limitations of the KNN algorithm in the local area.

Figure 6. The flow of data preprocessing: (a) Data blocking. (b) Construction of the neighbourhood system. (c) Neighbourhoods in structural and unstructured point clouds.

Figure 7. Extraction process of density features.

Figure 8. Density calculation method: (a) Build a 3D plane and project the point clouds. (b) The area of the minimum circumscribed rectangle between the plane point clouds and the projection point clouds is the range of density calculation.

Figure 9. Two feature fusion methods: (a) Feature interpolation based on the distance between adjacent point clouds. (b) Direct mapping method for feature fusion.

Figure 10. The flow of point segmentation.

Figure 11. The schematic diagram of blocks and neighbourhoods.

Table 2. Blocking details.

No.	Size of point spaces (m)	Size of blocks (m)	Number of blocks	Number of classes
1	1223.0 × 342.6 × 230.9	12.2 × 3.4 × 7.7	10352	9
2	1020.2 × 481.9 × 197.8	10.2 × 4.9 × 6.6	4487	9
3	1020.5 × 450.5 × 73.5	10.2 × 4.5 × 7.4	9788	9
4	1888.2 × 512.5 × 64.8	18.9 × 5.1 × 6.5	12453	9
5	887.1× 1336.1 × 205.3	8.9× 13.4 × 6.8	8964	9
6	1141.7 × 608.8× 64.8	11.4 × 6.1× 6.5	11036	9

Figure 12. The segmentation results of the power line and power tower point clouds: (a) Ground truth of No.5’s point clouds. (b) The results of No.5’s point clouds. (c) Ground truth of No.6’s point clouds. (d) The results of No.6’s point clouds. Green point clouds are power lines, blue point clouds are power towers and grey point clouds are other classes.

Figure 14. The trends of indicators for evaluating segmentation: (a) The trend of the mean IoU. (a2) The trend of the overall accuracy.

Table 3. Segmentation evaluation of different methods.

Method	Power line		Power tower		Mean IoU (%)	Overall accuracy (%)
	IoU (%)	Accuracy (%)	IoU (%)	Accuracy (%)
PointNet	51.47	60.82	45.84	49.76	42.51	50.63
PointCNN	77.26	85.72	75.31	87.43	75.89	86.81
Density + PointNet	65.41	72.54	50.32	56.87	51.72	64.59
Density + PointCNN	83.93	90.15	82.40	91.24	82.73	90.76
Point Transformer	79.84	87.31	77.53	88.86	78.52	88.15

Figure 15. The point clouds at different densities: (a) the original density; (b) 1/5th of the original density; (c) 1/10th of the original density; (d) 1/20th of the original density. Green point clouds are power lines, and blue point clouds are power towers.

Figure 16. The segmentation results at different densities (using the D + L mode): (a,e) the ground truth; (b,f) 1/5th of the original density; (c,g) 1/10th of the original density; (d,h) 1/20th of the original density. Green point clouds are power lines, and blue point clouds are power towers.

Table 4. Segmentation evaluation of different downsampling degrees (using D+L mode).

Downsampling degree	Power line		Power tower		Mean IoU (%)	Overall accuracy (%)
	IoU (%)	Accuracy (%)	IoU (%)	Accuracy (%)
1/5	80.21	87.41	77.15	88.23	79.43	88.05
1/10	83.93	90.15	82.40	91.24	82.73	90.76
1/20	76.43	82.53	76.56	83.64	74.01	81.22

Figure 17. The echo intensity of point clouds.

Figure 18. The influence of echo intensity on the segmentation results (using D + L mode): (a,d) the ground truth; (b,e) model without echo intensity; (c,f) model with echo intensity. Green point clouds are power lines, and blue point clouds are power towers.

Figure 19. The effect of echo intensity on model training time (using the D + L mode).

Table 5. Evaluation of the influence of echo intensity on segmentation (using the D + L mode).

	Power line		Power tower		Mean IoU (%)	Overall accuracy (%)
	IoU (%)	Accuracy (%)	IoU (%)	Accuracy (%)
Without intensity	83.93	90.15	82.40	91.24	82.73	90.76
With intensity	84.32	90.48	84.01	93.15	83.54	92.88

Data availability statement

The test point data used in this study are openly available in the Dryad Digital Repository at https://doi.org/10.5061/dryad.j3tx95xj1.