TLSNet: A semantic segmentation method for key point cloud regions of transmission lines.

Area of Science:

• Computer Vision
• Geospatial Data Analysis
• Electrical Engineering Infrastructure

Background:

• Current transmission line point cloud segmentation algorithms struggle with fine-grained structures and uneven data distribution.
• Accurate segmentation is crucial for digital operation, maintenance, and autonomous inspection of transmission lines.

Purpose of the Study:

• To develop an advanced point cloud segmentation model, TLSNet, specifically for transmission line infrastructure.
• To improve the accuracy and efficiency of segmenting complex transmission line point cloud data.

Main Methods:

• Proposed TLSNet model incorporating a dynamic density adaptation mechanism (DALNF-Layer) for uneven point cloud distribution.
• Utilized a hierarchical-offset transformer (DLCTransformer) for cross-scale context modeling.
• Implemented an inverted residual module (InvResMLP) to optimize gradient propagation.

Main Results:

• TLSNet achieved precise segmentation of key regions in transmission line point clouds.
• The proposed model demonstrated superior segmentation accuracy compared to existing algorithms.
• The framework proved efficient for complex transmission line point cloud scenarios.

Conclusions:

• TLSNet offers a significant advancement in transmission line point cloud segmentation.
• The model provides a robust technical foundation for digital transmission line management and autonomous UAV inspection.
• This research facilitates enhanced operational efficiency and safety in power infrastructure maintenance.