成功大學博碩士論文系統

簡易檢索 / 詳目顯示

回結果列表

研究生：	陳彥之 Chen, Yen-Chih
論文名稱：	基於局部高程注意力 Point Transformer 應用於空載光達地面點萃取 Point Transformer with Local Elevation Attention for Ground Point Extraction from Airborne LiDAR Data
指導教授：	林昭宏 Lin, Chao-Hung
學位類別：	碩士 Master
系所名稱：	工學院 - 測量及空間資訊學系 Department of Geomatics
論文出版年：	2026
畢業學年度：	114
語文別：	英文
論文頁數：	70
中文關鍵詞：	深度學習、地面點分類、空載光達
外文關鍵詞：	Deep Learning, Ground Point Classification, Airborne LiDAR
相關次數：	點閱：103 下載：1
分享至:	分享至facebook 分享至twitter

查詢本校圖書館目錄查詢臺灣博碩士論文知識加值系統勘誤回報

數值地形模型（Digital Elevation Model, DEM）為描述地表的重要資料，亦為國家級基礎圖資之一，廣泛應用於災害防治、工程規劃與土地管理等領域。DEM 可透過多種技術生成，包括攝影測量、干涉式合成孔徑雷達（Interferometric Synthetic Aperture Radar, InSAR）以及光達（Light Detection and Ranging, LiDAR）。其中，空載光達藉由發射雷射脈衝並量測其回波時間，可取得高解析度且高精度的地形點雲資料，特別適合用於大尺度地形測繪。然而，空載光達點雲資料具有資料量龐大且結構不規則的特性，使得後續資料處理面臨挑戰，尤其是在地面點分類任務中，傳統方法多仰賴人工操作或規則式濾波，流程繁瑣且耗時。為提升地面點分類之效率與穩定性，本研究透過深度學習方法，並使用臺灣 1:5000 空載光達 DEM 資料作為資料集，涵蓋 2016 至 2020 年間中南部地區之實測成果。所提出的方法以 Point Transformer 架構為基礎，利用自注意力機制直接建模不規則三維點雲中的局部幾何關係。此外，本研究進一步提出一套局部高程注意力（Local Elevation Attention）模組，於初始特徵擷取階段編碼局部高程差異，以強化模型對高度變化的感知能力。透過將高程敏感的注意力機制引入特徵聚合流程，所提出的方法能更有效區分地面點與非地面點，並適用於多樣化的地形環境。實驗結果顯示，該方法在都市、農田與森林等不同地形類型下皆能維持穩定且一致的分類表現，並能有效處理複雜地形結構與大規模點雲資料，驗證了結合局部高程注意力之 Point Transformer 架構於不同地形條件下的可行性與穩定表現。

Digital Elevation Models (DEMs) are fundamental representations of terrain information and serve as essential national base maps, with wide applications in disaster prevention, construction planning, and land management. DEMs can be generated using various techniques, including photogrammetry, interferometric synthetic aperture radar (InSAR), and Light Detection and Ranging (LiDAR). Among these approaches, airborne LiDAR provides high-resolution and high-accuracy terrain measurements by emitting laser pulses and recording their return times, making it particularly suitable for large-scale topographic mapping. However, the massive volume and irregular structure of airborne LiDAR point clouds pose significant challenges for data processing, especially for ground point classification, which traditionally relies on manual or rule-based methods and is therefore labor-intensive and time-consuming. To improve the efficiency and robustness of ground point classification, this study adopts a deep learning based approach using airborne LiDAR data from Taiwan’s 1:5000 DEM dataset, covering surveyed areas in central and southern Taiwan from 2016 to 2020. The proposed method is built upon the Point Transformer architecture, which leverages self-attention mechanisms to directly model local geometric relationships within irregular 3D point clouds. Furthermore, a Local Elevation Attention module is integrated into the network to explicitly encode local elevation differences and enhance height-aware feature representation. By incorporating elevation-sensitive attention at the early feature extraction stage, the proposed approach improves the discrimination between ground and non-ground points across diverse terrain types. Experimental results demonstrate that the proposed method achieves stable and consistent performance in urban, farmland, and forest environments, while effectively handling complex terrain structures and large-scale point clouds. These results validate the feasibility and robustness of integrating Local Elevation Attention into the Point Transformer architecture for airborne LiDAR ground point classification under diverse terrains.

摘要 i
ABSTRACT ii
ACKNOWLEDGEMENTS iii
TABLE OF CONTENTS iv
LIST OF TABLES vi
LIST OF FIGURES vii

CHAPTER 1 Introduction 1
1 Background 1
2 Motivation and Objective 2
3 Contributions 3
4 Structure of the Thesis 4

CHAPTER 2 Related Works 5
1 Traditional ALS Point Cloud Filtering Methods 5
1.1 Slope-Based Filter 5
1.2 Progressive Morphological Filter (PMF) 6
1.3 Adaptive TIN Model 6
2 Deep Learning Applications in ALS Point Cloud Filtering 7
2.1 Projection-Based Deep Learning Methods for ALS DTM Generation 8
2.2 Feature-Based Deep Learning Methods for ALS Ground Point Filtering 8
2.3 Ensemble Learning for ALS Point Cloud Classification 9
3 Deep Learning Architectures for 3D Point Cloud 10
3.1 Image-based Methods 11
3.2 Voxel-based Methods 13
3.3 Point-based Methods 15

CHAPTER 3 Methodology 18
1 Motivation for Transformer 19
1.1 Transformer Architecture 19
1.2 Vector Attention 23
2 Core Algorithmic Design of the Point Transformer 25
2.1 Attention Mechanism 25
2.2 Positional Encoding 26
3 Integration of Local Elevation Attention into the Point Transformer 27
3.1 Model Architecture 27
3.2 Proposed Local Elevation Attention Module 30

CHAPTER 4 Experimental Results and Discussion 33
1 Dataset 33
2 Data Preprocessing 35
2.1 Patch partition 35
2.2 Downsampling 36
3 Inference Strategy 38
3.1 Stage 1: Index-Based Sampling 39
3.2 Stage 2: Nearest-Neighbor Batch Sampling 39
4 Experimental Results 40
4.1 Experimental Setup 41
4.2 Performance Evaluation 41
4.3 Results Visualization 44
4.4 Error Analysis and Discussion 52

CHAPTER 5 Conclusion 55
1 Conclusions 55
2 Future Work 56

REFERENCES 57
                                    

Arif, F., Maulud, K. N. A., and Rahman, A. A. A. (2018). Generation of digital elevation model through aerial technique. In IOP Conference Series: Earth and Environmental Science, volume 169, page 012093. IOP Publishing. IGRSM 2018.

Axelsson, P. (2000). Dem generation from laser scanner data using adaptive tin models. In International Archives of Photogrammetry and Remote Sensing, volume XXXIII, Part B4, pages 110–117.

Ciou, T.-S., Lin, C.-H., and Wang, C.-K. (2024). Airborne lidar point cloud classification using ensemble learning for dem generation. Sensors, 24:6858.

Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., Uszkoreit, J., and Houlsby, N. (2020). An image is worth 16x16 words: Transformers for image recognition at scale. CoRR, abs/2010.11929.

Graham, B., Engelcke, M., and van der Maaten, L. (2018). 3d semantic segmentation with submanifold sparse convolutional networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 9224–9232. IEEE.

Hu, J. and Li, J. (2024). A brief discussion on the overall classification algorithm of airborne lidar point cloud. In ISPRS TC I Mid-term Symposium “Intelligent Sensing and Remote Sensing Application＂, volume XLVIII-1-2024 of The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, pages 217–220, Changsha, China. This contribution has been peer-reviewed. © Author(s) 2024. CC BY 4.0 License.

Hu, X. and Yuan, Y. (2016). Deep-learning-based classification for dtm extraction from als point cloud. Remote Sensing, 8.

Huang, S., Hu, Q., Zhao, P., Li, J., Ai, M., and Wang, S. (2023). Als point cloud semantic segmentation based on graph convolution and transformer with elevation attention. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, PP:1–14.

Julzarika, A. and Djurdjani, D. (2019). Dem classifications: opportunities and potential of its applications. Journal of Degraded and Mining Lands Management, 6:1897–1905.

Li, L., Shum, H. P. H., and Breckon, T. P. (2023). Less is more: Reducing task and model complexity for 3d point cloud semantic segmentation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 9361–9370. IEEE.

Liu, X., Hu, H., and Hu, P. (2015). Accuracy assessment of lidar-derived digital elevation models based on approximation theory. Remote Sensing, 7:7062–7079.

Maturana, D. and Scherer, S. (2015). Voxnet: A 3d convolutional neural network for real-time object recognition. In 2015 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 922–928.

Milioto, A., Vizzo, I., Behley, J., and Stachniss, C. (2019). Rangenet ++: Fast and accurate lidar semantic segmentation. In 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 4213–4220.

Nurunnabi, A., Teferle, F. N., Li, J., Lindenbergh, R. C., and Hunegnaw, A. (2021). An efficient deep learning approach for ground point filtering in aerial laser scanning point clouds. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, XLIII-B1-2021:31–38.

Qi, C. R., Su, H., Mo, K., and Guibas, L. J. (2017). Pointnet: Deep learning on point sets for 3d classification and segmentation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 652–660. IEEE.

Shamsi, U. (2001). Dem applications in hydrologic modeling. Journal of Water Management Modeling, R207-12. Formerly in Models and Applications to Urban Water Systems. ISBN: 0-9683681-4-X.

Toz, G. and Erdogan, M. (2008). Dem (digital elevation model) production and accuracy modeling of dems from 1:35,000 scale aerial photographs. In The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, volume XXXVII, Beijing, China. ISPRS.

Van Nieuwenhuizen, N., Lindsay, J. B., and DeVries, B. (2021). Automated mapping of transportation embankments in fine-resolution lidar dems. Remote Sensing, 13(7):1308.

Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L. u., and Polosukhin, I. (2017). Attention is all you need. In Guyon, I., Luxburg, U. V., Bengio, S., Wallach, H., Fergus, R., Vishwanathan, S., and Garnett, R., editors, Advances in Neural Information Processing Systems, volume 30. Curran Associates, Inc.

Vosselman, G. (2000). Slope based filtering of laser altimetry data. IAPRS, XXXIII.

Wang, Y., Sun, Y., Liu, Z., Sarma, S. E., Bronstein, M. M., and Solomon, J. M. (2019). Dynamic graph cnn for learning on point clouds. ACM Transactions on Graphics (TOG), 38(5):146:1–146:12.

Wen, C., Li, X., Yao, X., Peng, L., and Chi, T. (2021). Airborne lidar point cloud classification with global-local graph attention convolution neural network. ISPRS Journal of Photogrammetry and Remote Sensing, 173:181–194.

Wu, B., Wan, A., Yue, X., and Keutzer, K. (2018). Squeezeseg: Convolutional neural nets with recurrent crf for real-time road-object segmentation from 3d lidar point cloud. In 2018 IEEE International Conference on Robotics and Automation (ICRA), pages 1887–1893.

Yuan, F., Zhang, J. X., Zhang, L., and Gao, J. X. (2009). Dem generation from airborne lidar data. In Proceedings of the International Conference on Geoinformatics and Spatial Environmental Modeling (GSEM), volume XXXVIII-7/C4 of The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, pages 308–312, Beijing, China.

Zhang, J., Lin, X., and Ning, X. (2013). Svm-based classification of segmented airborne lidar point clouds in urban areas. Remote Sensing, vol. 5, issue 8, pp. 3749-3775, 5:3749–3775.

Zhang, K., Chen, S.-C., Whitman, D., Shyu, M.-L., Yan, J., and Zhang, C. (2003). A progressive morphological filter for removing nonground measurements from airborne lidar data. IEEE Transactions on Geoscience and Remote Sensing, 41(4):872–882.

Zhang, W., Qi, J., Wan, P., Wang, H., Xie, D., Wang, X., and Yan, G. (2016). An easy-to-use airborne lidar data filtering method based on cloth simulation. Remote Sensing, 8(6):501. Received: 13 March 2016; Accepted: 3 June 2016; Published: 15 June 2016.

Zhao, H., Jiang, L., Jia, J., Torr, P. H. S., and Koltun, V. (2021). Point transformer. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 16259–16268.

Zhao, Y., Mou, L., and Zhu, X. X. (2020). Exploring self-attention for image recognition. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 10076–10085.

校外：立即公開

簡易檢索 / 詳目顯示

相關論文