Semantic Segmentation of Large‑Scale Laser Point Cloud in Mines Based on Local Feature Enhancement

Fig. 11. Comparison of semantic segmentation details for mining point clouds. (a),(b) Semantic segmentation results of scenario 1 before and after removing spatial position encoding unit; (c),(d) semantic segmentation results of scenario 2 before and after removing spatial position encoding unit; (e),(f) semantic segmentation results of scenario 3 before and after removing spatial normal vector encoding units

Download full size

Table 1. Semantic segmentation results on mining point cloud datasets

View table

Table 1. Semantic segmentation results on mining point cloud datasets

Method	OA	mIoU	IoU
Method	OA	mIoU	Noise	Ground	Barricade	Hillside	Truck
PointNet	96.20	79.610	84.50	96.70	79.50	82.50	52.60
PointNet++	91.84	64.260	68.70	92.90	63.80	69.70	26.20
PointCNN	96.92	75.350	87.59	97.09	82.10	86.44	23.53
RandLA-Net	97.30	85.818	91.00	97.26	86.48	89.12	65.23
SCF-Net	97.10	82.238	89.97	96.69	84.22	81.38	57.93
BAF-LAC^［23］	97.50	86.380	91.06	97.47	87.76	89.15	66.46
DLA-Net^［24］	98.40	86.854	91.32	97.24	86.45	89.13	70.16
LFE-Net	97.92	87.578	90.98	97.74	88.71	89.53	70.93

Table 2. Computational efficiency and parameters on mining point cloud datasets
View table
Table 2. Computational efficiency and parameters on mining point cloud datasets
Method Total inference time /s Parameters /10⁶ mIoU /%
RandLA-Net 344.07 4.99 85.818
SCF-Net 353.63 12.15 82.238
BAF-LAC 412.46 11.64 86.380
DLA-Net 367.56 5.07 86.854
LFE-Net 357.99 7.35 87.578

Table 3. LFE-Net ablation experiment results

View table

Table 3. LFE-Net ablation experiment results

	mIoU	IoU
	mIoU	Noise	Ground	Barricade	Hillside	Truck
1） Removing spatial position encoding unit	84.842	89.60	96.23	84.39	86.62	67.37
2） Removing spatial normal vector encoding unit	85.286	90.51	97.20	86.26	87.87	64.59
3） Removing semantic feature encoding unit	86.490	90.95	97.51	87.99	88.46	68.54
4） Removing mixed pooling	86.596	90.62	97.34	86.82	89.46	68.74
5） LFE-Net	87.578	90.98	97.74	88.71	89.53	70.93

Table 4. Normal vector encoding ablation experiment unit:%
View table
Table 4. Normal vector encoding ablation experiment unit:%
Spatial normal vector encoding mIoU
1） $(n_{i} \oplus n_{i j})$ 86.604
2） $(n_{i} - n_{i j})$ 87.578
3） $(n_{i} \oplus n_{i j} \oplus (n_{i} - n_{i j}))$ 85.618

Tools

Get Citation

Copy Citation Text

Hongxiang Dong, Yi An, Lirong Xie, Zhiyong Yang, Kai Zhang. Semantic Segmentation of Large‑Scale Laser Point Cloud in Mines Based on Local Feature Enhancement[J]. Chinese Journal of Lasers, 2024, 51(17): 1710002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites