Infrared and Laser Engineering, Volume. 52, Issue 11, 20230212(2023)
Airborne LiDAR data classification method combining physical and geometric characteristics
Figures & Tables(12)
Fig. 1. Different LiDAR irradiation situations. (a) Plane perpendicular to LiDAR beam; (b) Inclined plane; (c) Multiple targets
Fig. 2. Structure of FW and FF block. (a) Structure of FW block; (b) Structure of FF block
Fig. 4. Local neighborhood feature enhancement. (a) Local neighborhood fully connected structure; (b) Structure of LE block
Fig. 6. Full-waveform of sampling points of different categories. (a) Ground full-waveform; (b) Vegetation full-waveform; (c) Building full-waveform; (d) Power line full-waveform; (e) Transmission tower full-waveform; (f) Street path full-waveform
Fig. 7. Visualization of classification results. (a) Input data; (b) Classification results of FWNet2; (c) Classification results of proposed method; (d) Real category label
Table 1. Dataset distribution
View tableView in ArticleTable 1. Dataset distribution
Label Class Train Test Number Number 1 Ground 1787352 20.4% 193070 18.1% 2 Vegetation 4719634 53.9% 765327 71.7% 3 Building 1514486 17.3% 49138 4.6% 4 Power line 71978 0.8% 8152 0.8% 5 Transmission tower 32008 0.4% 1829 0.2% 6 Street path 633606 7.2% 49580 4.6%
Table 2. Quantitative comparison of different methods
View tableView in ArticleTable 2. Quantitative comparison of different methods
Method Metric Ground Vegetation Building Power line Transmission tower Street path Mean CNN[ 14 ]Precision 0.60 0.96 0.07 0.10 0.01 0.18 0.32 Recall 0.18 0.78 0.11 0.88 0.44 0.60 0.50 F1 score 0.28 0.86 0.09 0.19 0.02 0.28 0.28 FCN[ 14 ]Precision 0.79 0.95 0.99 0.98 0.63 0.67 0.84 Recall 0.84 0.97 0.93 0.91 0.88 0.69 0.87 F1 score 0.81 0.96 0.96 0.94 0.73 0.68 0.85 PointNet[ 9 ]Precision 0.56 0.97 0.95 0.92 0.61 0.94 0.83 Recall 0.91 0.85 0.83 0.84 0.48 0.62 0.76 F1 score 0.69 0.91 0.88 0.88 0.53 0.75 0.77 PointNet++[ 10 ]Precision 0.56 0.98 0.99 0.97 0.52 0.96 0.83 Recall 0.93 0.85 0.85 0.85 0.66 0.60 0.83 F1 score 0.70 0.91 0.91 0.91 0.58 0.74 0.80 GFSAE[ 13 ]Precision 0.77 0.97 0.93 0.96 0.70 0.83 0.86 Recall 0.90 0.95 0.93 0.93 0.73 0.62 0.84 F1 score 0.83 0.96 0.93 0.94 0.71 0.71 0.85 RandLA-Net[ 11 ]Precision 0.74 0.98 0.83 0.93 0.81 0.99 0.88 Recall 0.95 0.93 0.95 0.97 0.67 0.60 0.85 F1 score 0.83 0.95 0.89 0.95 0.73 0.75 0.85 GACNN[ 12 ]Precision 0.76 0.98 0.87 0.95 0.74 0.94 0.87 Recall 0.95 0.94 0.95 0.98 0.79 0.63 0.87 F1 score 0.84 0.96 0.91 0.96 0.76 0.75 0.86 FWNet2[ 16 ]Precision 0.80 0.98 0.99 0.99 0.99 0.97 0.95 Recall 0.94 0.96 0.95 0.99 0.79 0.71 0.89 F1 score 0.86 0.97 0.97 0.99 0.88 0.82 0.92 Proposed method Precision 0.83 0.98 0.98 0.99 0.97 0.98 0.96 Recall 0.95 0.96 0.95 0.99 0.82 0.75 0.90 F1 score 0.89 0.97 0.96 0.99 0.89 0.85 0.92
Table 3. Ablation experiment on FF block
View tableView in ArticleTable 3. Ablation experiment on FF block
Model FW block Fusion Mean precision Mean recall Mean F1 score A × $\oplus $ 0.90 0.86 0.88 B √ $\oplus $ 0.92 0.88 0.89 C × √ 0.94 0.89 0.90 Proposed method √ √ 0.96 0.90 0.92
Table 4. Ablation experiment on stacking layers of FW block
View tableView in ArticleTable 4. Ablation experiment on stacking layers of FW block
Model Number of layers Mean precision Mean recall MeanF1 score D 1 0.94 0.89 0.91 E 2 0.95 0.90 0.91 F 3 0.96 0.90 0.92 G 4 0.95 0.89 0.91
Table 5. Ablation experiments on LE and GE block
View tableView in ArticleTable 5. Ablation experiments on LE and GE block
Model LE GE Mean precision Mean recall Mean F1 score LI block Attention Pooling H MLP Max pooling √ 0.91 0.86 0.88 I √ Max pooling √ 0.94 0.88 0.89 J MLP √ √ 0.92 0.88 0.89 K √ √ × 0.94 0.88 0.90 Proposed method √ √ √ 0.96 0.90 0.92
Tools
Get Citation
Copy Citation Text
Yiqiang Zhao, Qi Zhang, Changlong Liu, Weikang Wu, Yao Li. Airborne LiDAR data classification method combining physical and geometric characteristics[J]. Infrared and Laser Engineering, 2023, 52(11): 20230212
Paper Information
Category:
Received: Apr. 10, 2023
Accepted: --
Published Online: Jan. 8, 2024
The Author Email: Li Yao (liyao@tju.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20