Comparison of machine learning algorithm performances in digital terrain model generation

Öz

LiDAR technology enables precise distance measurements by emitting laser pulses that reflect off surface objects, allowing for the calculation of spatial coordinates. Alongside spatial data associated color values of LiDAR points can be extracted from images captured by onboard cameras. As the laser beams reflect upon their initial contact with surfaces, the resulting point cloud must be appropriately classified to support specific analytical or operational objectives. This study uses different machine learning methods to sort and label LiDAR point cloud data into ground and non-ground points, then compares how well each method works. For this purpose, a dataset acquired by an unmanned aerial vehicle over the Democratic Republic of Congo was utilized. The dataset comprises 114,557 points, each described by three geometric features (DeltaH, Verticality, 3rd Eigenvalue) and two normalized color attributes (Red and Green Ratios), derived from RGB values. A total of ten machine learning algorithms were implemented and assessed. Among them, the XGBoost algorithm demonstrated the highest classification accuracy at 84.1%, while the Naive Bayes algorithm yielded the lowest accuracy, at 72.4%.

Anahtar Kelimeler

• Remote sensing
• LiDAR
• Photogrammetry
• Machine learning
• Classification

Sayısal arazi modeli oluşturmada makine öğrenme algoritma performanslarının karşılaştırılması

Öz

LiDAR teknolojisi, yüzey nesnelerinden yansıyan lazer darbeleri göndererek hassas mesafe ölçümleri yapılmasına olanak tanır ve bu sayede mekânsal koordinatların hesaplanması mümkün olur. Mekânsal verilerin yanı sıra, LiDAR noktalarına ait renk bilgileri de araç üzerindeki kameralarla çekilen görüntülerden elde edilebilir. Lazer ışınları yüzeylerle ilk temas ettikleri anda yansıdığından, ortaya çıkan nokta bulutunun belirli analizsel veya operasyonel amaçlara hizmet edebilmesi için uygun şekilde sınıflandırılması gerekmektedir. Bu çalışmada, LiDAR nokta bulutu verilerini sıralamak ve analiz etmek için çeşitli makine öğrenmesi yöntemleri kullanılmış ve her bir yöntemin performansı karşılaştırılmıştır. Bu amaçla, insansız hava aracı ile Demokratik Kongo Cumhuriyeti’nde elde edilen bir veri seti kullanılmıştır. Veri seti, üç geometrik özellik ve iki renk bilgisi içeren toplam 114 557 noktadan oluşmaktadır. On farklı makine öğrenmesi algoritması uygulanmış ve değerlendirilmiştir. Bu algoritmalar arasında XGBoost, %84.1 ile en yüksek sınıflandırma doğruluğunu gösterirken, Naive Bayes algoritması ile %72.4 ile en düşük doğruluğa ulaşılmıştır.

Anahtar Kelimeler

• Uzaktan algılama
• LiDAR
• Fotogrametri
• Makine öğrenmesi
• Sınıflandırma

Kaynakça

1. Almohsen, A. S. (2024). Challenges facing the use of remote sensing technologies in the construction industry: A review. Buildings, 14(9), 2861.
2. Bishop, C. M. (1995). Neural networks for pattern recognition. Oxford University Press.
3. Blomley, R., Weinmann, M., Leitloff, J., & Jutzi, B. (2014). Shape distribution features for point cloud analysis–A geometric histogram approach on multiple scales. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 2, 9-16.
4. Breiman, L. (1996). Bagging predictors. Machine Learning, 24(2), 123-140.
5. Breiman, L. (2001). Random forests. Machine Learning, 45(1), 5-32.
6. Breiman, L., Friedman, J. H., Olshen, R. A., & Stone, C. J. (1984). Classification and regression trees. Wadsworth International Group.
7. Chen, T., & Guestrin, C. (2016, August). Xgboost: A scalable tree boosting system. In Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining (pp. 785-794). San Francisco, California, USA.
8. Cortes, C., & Vapnik, V. (1995). Support-vector networks. Machine Learning, 20(3), 273-297.

9. Cover, T., & Hart, P. (1967). Nearest neighbor pattern classification. IEEE Transactions on Information Theory, 13(1), 21-27.
10. Duran, Z., Ozcan, K., & Atik, M. E. (2021). Classification of photogrammetric and airborne LiDAR point clouds using machine learning algorithms. Drones, 5(4), 104.
11. Gharineiat, Z., Tarsha Kurdi, F., & Campbell, G. (2022). Review of automatic processing of topography and surface feature identification LiDAR data using machine learning techniques. Remote Sensing, 14(19), 4685.
12. Hosmer, D. W., & Lemeshow, S. (2000). Applied logistic regression (2nd ed.). New York: Wiley.
13. Jakovljevic, G., Govedarica, M., Alvarez-Taboada, F., & Pajic, V. (2019). Accuracy assessment of deep learning-based classification of LiDAR and UAV point clouds for DTM creation and flood risk mapping. Geosciences, 9(7), 323.
14. Joseph-Rivlin, M., Zvirin, A., & Kimmel, R. (2019, October). Moment: Flavor the moments in learning to classify shapes. In Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops (ICCVW) (pp. 4085-4094). Seoul, South Korea.
15. Kang, Z., Yang, J., & Zhong, R. (2017). A Bayesian-network-based classification method integrating airborne LiDAR data with optical images. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 10(4), 1651-1661.
16. Kuçak, R. A. (2022, June). The analysis of 3D geometric features on point clouds by using open-source software. In Proceedings of the 4th Intercontinental Geoinformation Days (pp. 87-90). Tabriz, Iran.
17. Liu, X., Zhang, Z., Peterson, J., & Chandra, S. (2007, December). The effect of LiDAR data density on DEM accuracy. In Proceedings of the 17th International Congress on Modelling and Simulation (MODSIM07) (pp. 1363-1369). Christchurch, New Zealand: Modelling and Simulation Society of Australia and New Zealand.
18. Maturana, D., & Scherer, S. (2015, May). 3D convolutional neural network for landing zone detection from LiDAR. In Proceedings of the IEEE International Conference on Robotics and Automation (ICRA) (pp. 3471-3478). Seattle, Washington, USA.
19. McCallum, A., & Nigam, K. (1998, July). A comparison of event models for naive bayes text classification. In AAAI-98 workshop on learning for text categorization (Vol. 752, No. 1, pp. 41-48). Madison, Wisconsin, USA.
20. Özdemir, E., Remondino, F., & Golkar, A. (2019). Aerial point cloud classification with deep learning and machine learning algorithms. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, XLII-4/W18, 843-849.
21. Özen, A. C., & Çilengiroğlu, Ö. V. (2024, May). Performance comparison of classification methods in LiDAR point clouds. In Proceedings of the International Applied Statistics Congress (UYIK). Tokat, Türkiye.
22. Park, Y., & Guldmann, J.-M. (2019). Creating 3D city models with building footprints and LiDAR point cloud classification: A machine learning approach. Computers, Environment and Urban Systems, 75, 76-89.
23. Schmidhuber, J. (2015). Deep learning in neural networks: An overview. Neural Networks, 61, 85-117.
24. Tan, Y., Liu, X., Shuaishuai, J., Wang, Q., Wang, D., & Xie, X. (2023). A terrestrial laser scanning-based method for indoor geometric quality measurement. Remote Sensing, 16(1), 59.
25. Teruggi, S., Grilli, E., Russo, M., Fassi, F., & Remondino, F. (2020). A hierarchical machine learning approach for multi-level and multi-resolution 3D point cloud classification. Remote Sensing, 12(16), 2598.
26. Wu, L., Zhu, X., Lawes, R., Dunkerley, D., & Zhang, H. (2019). Comparison of machine learning algorithms for classification of LiDAR points for characterization of canola canopy structure. International Journal of Remote Sensing, 40(15), 5973-5991.
27. URL-1: Kılıç, İ. Principal component analysis (PCA): A practical guide. Medium. https://medium.com/@ilyurek/principal-component-analysis-pca-a-practical-guide-58dea99dd93 (Accessed: 18 January 2025).
28. URL-2: Yellowscan. How does LiDAR work? Yellowscan Knowledge Base. https://www.yellowscan.com/knowledge/how-does-LiDAR-work (Accessed: 18 January 2025).
29. URL-3: Wasser, L. A. What is a CHM, DSM and DTM? About gridded, raster LiDAR data. NEON Science. https://www.neonscience.org/resources/learning-hub/tutorials/chm-dsm-dtm (Accessed: 18 January 2025).