Land Cover Classification Based on Airborne Lidar Point Cloud with Possibility Method and Multi-Classifier

Area of Science:

• Geospatial data analysis
• Remote sensing
• Urban planning

Background:

• Aerial laser scanner (ALS) point clouds offer crucial 3D data for urban land cover studies.
• Current single-classifier methods struggle with accuracy in ambiguous areas due to decision uncertainty.

Purpose of the Study:

• To enhance point cloud classification accuracy by reducing uncertainty in decision-making.
• To develop a multi-classifier fusion method leveraging possibility theory.

Main Methods:

• Feature importance analysis using XGBoost to define a feature space.
• Utilizing two Support Vector Machines (SVMs) as base classifiers.
• Quantitatively evaluating classifier outputs to identify confusing areas and calculating confidence weights.

Main Results:

• The proposed method successfully identifies and addresses classification uncertainties in ambiguous regions.
• Multi-classifier fusion based on possibility theory significantly boosts classification accuracy.
• Validation on DALES datasets confirms the method's effectiveness.

Conclusions:

• The developed possibility theory-based fusion method offers a robust solution for improving urban land cover classification from ALS point clouds.
• This approach enhances the reliability of geospatial data for digital city initiatives.