Airborne LiDAR Point Cloud Classification Using Ensemble Learning for DEM Generation.

Area of Science:

• Geospatial science and remote sensing
• Computer science, specifically machine learning and artificial intelligence

Background:

• Airborne laser scanning (ALS) point clouds are crucial for digital elevation model (DEM) generation.
• Traditional DEM generation involves complex point cloud classification and manual error correction.
• Existing deep learning models struggle with diverse terrains due to training on simplified data.

Purpose of the Study:

• To develop a robust point-based deep learning model for accurate ground point classification in ALS data.
• To enhance the quality and accuracy of DEMs generated from challenging terrains.
• To improve the efficiency of DEM generation by reducing manual post-processing.

Main Methods:

• Proposed a point-based deep learning model incorporating boosting ensemble learning.
• Utilized a set of geometric features as input for the model.
• Integrated specialized ground point classifiers tailored for different terrain types within the ensemble strategy.

Main Results:

• Achieved significant improvements in point cloud classification accuracy (from 80.9% to 92.2%) and F1 score (from 82.2% to 94.2%).
• Reduced DEM generation error (RMSE) from 0.318-1.362 m to 0.273-1.032 m across various terrains.
• Demonstrated enhanced classification robustness and accuracy on diverse terrain datasets.

Conclusions:

• The proposed ensemble learning approach effectively boosts the performance of deep learning models for ALS point cloud classification.
• The method significantly improves the quality of generated DEMs, especially in complex geographical areas.
• This approach offers a more accurate and efficient alternative to traditional DEM generation techniques.