A Method Considering Multi-Dimensional Feature Differences for Extracting Rural Buildings Based on Airborne LiDAR.

Area of Science:

• Geospatial Science
• Remote Sensing
• Computer Vision

Background:

• Extracting building data from airborne point clouds is challenging, especially in rural areas with dense, similar-height vegetation.
• Existing methods often struggle with complex rural environments, necessitating specialized approaches.

Purpose of the Study:

• To develop and validate an effective building classification procedure for airborne point clouds in typical rural Chinese settings.
• To address the scarcity of research on building extraction in complex rural environments.

Main Methods:

• Dynamic multi-level grid segmentation based on slope analysis to identify terrain types.
• Differentiated filtering parameters for ground point extraction tailored to various terrains.
• Watershed segmentation and geometric feature analysis to select building regions of interest.
• Refined classification using morphological differences between buildings and other objects.

Main Results:

• Achieved high classification performance with precision, recall, and F1 scores exceeding 93.37%, 97.05%, and 95.17% on test datasets.
• Demonstrated average precision, recall, and F1 scores of 94.02%, 97.20%, and 95.58%, respectively.
• Successfully classified buildings in complex rural environments, proving the method's adaptability.

Conclusions:

• The proposed method effectively extracts building information from airborne point clouds in challenging rural areas.
• The approach shows strong adaptability and practicality for diverse building data extraction tasks.
• This research contributes a robust solution for building classification in complex geospatial scenarios.