Filtering-Assisted Airborne Point Cloud Semantic Segmentation for Transmission Lines.

Area of Science:

• Geospatial Engineering
• Computer Vision
• Machine Learning

Background:

• Point cloud semantic segmentation is vital for transmission line analysis.
• Current methods struggle with large datasets, complex scenes, and imbalanced data in transmission line scenarios.
• Efficient and accurate segmentation is needed for infrastructure monitoring.

Purpose of the Study:

• To propose a filter-assisted semantic segmentation method for airborne point clouds of transmission lines.
• To address challenges of data volume, scene complexity, and sample imbalance.
• To enhance the efficiency and accuracy of transmission line point cloud analysis.

Main Methods:

• Introduced a cloth simulation filter to identify ground point clouds, reducing sample imbalance.
• Defined multi-dimensional features for classification.
• Trained a classification model for multi-element semantic segmentation of transmission line scenes.

Main Results:

• The filter-assisted algorithm significantly improved transmission line point cloud semantic segmentation.
• Achieved over 25.46% enhancement in segmentation efficiency and 3.15% in accuracy.
• Reduced data volume, sample classes, and imbalance index, decreasing processing time.

Conclusions:

• The proposed method offers a practical solution for efficient and accurate transmission line point cloud segmentation.
• It demonstrates significant theoretical and engineering value for scene reconstruction and intelligent understanding.
• The approach effectively mitigates challenges posed by large-scale, complex, and imbalanced point cloud data.