RSA-PT: A Point Transformer-Based Semantic Segmentation Network for Uninterrupted Operation in a Distribution Network...

Area of Science:

• Electrical Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Digitization of distribution networks is vital for quality of life and economic growth.
• Point cloud technology is key to intelligent transformation in distribution networks.
• Uninterrupted operation requires accurate scene understanding.

Purpose of the Study:

• To propose a deep learning network, RSA-PT, for semantic segmentation of distribution network point clouds.
• To improve the accuracy and efficiency of point cloud segmentation for critical infrastructure.
• To provide a technical foundation for digital analysis of distribution networks.

Main Methods:

• Developed RSA-PT, a deep learning network incorporating an improved residual spatial attention (RSA) module.
• Modified the network's loss function for enhanced performance.
• Segmented point clouds into ten classes relevant to distribution networks (e.g., lines, towers, ground, obstacles).

Main Results:

• RSA-PT achieved high performance metrics: 90.55% mean intersection over union (mIoU), 94.20% mean accuracy (mA), and 97.20% overall accuracy (OA).
• The proposed RSA-PT model demonstrated a significant improvement in mIoU, exceeding the baseline model by 6.63%.
• Ablation studies and model comparisons validated the effectiveness of the RSA module and network modifications.

Conclusions:

• The RSA-PT network effectively performs semantic segmentation of distribution network point clouds.
• This technology supports the digital analysis required for ensuring uninterrupted power distribution.
• The findings contribute to advancing the intelligent transformation of power distribution systems.