Title: PointNet++ model based on directional attention for insulators segmentation
Authors: Shucai Li; Enliang Chu; Pengcheng Wang; Xi Gao; Na Zhang
Addresses: State Grid Shanxi Lvliang Power Supply Company, Lvliang, Shanxi, 033000, China ' State Grid Shanxi Lvliang Power Supply Company, Lvliang, Shanxi, 033000, China ' State Grid Shanxi Lvliang Power Supply Company, Lvliang, Shanxi, 033000, China ' State Grid Shanxi Lvliang Power Supply Company, Lvliang, Shanxi, 033000, China ' State Grid Shanxi Electric Power Research Institute, Taiyuan, Shanxi, 030001, China
Abstract: Insulators are crucial components for the safety and reliability of power systems. However, due to their small size and complex structure, precise segmentation of insulators is challenging. To address this issue, this paper proposes a directional attention-based PointNet++ model (PDA). The core module of PDA is the directional attention (DA) module, which consists of spatial self-attention (SSA) and channel self-attention (CSA). This module is designed to establish long-range relationships in both spatial and channel directions of the feature map, enabling global modelling. Additionally, to reduce computational costs, multi-scale pyramid pooling is embedded in both the SSA and CSA modules. Notably, by integrating DA into PointNet++, the model enhances the correlation between point cloud features and the long-range dependency of positional information without significantly increasing the computational burden. Experimental results demonstrate that the PDA model significantly outperforms existing models in segmenting insulator point clouds from multiple power transmission corridors.
Keywords: PointNet++; PDA; spatial self-attention; SSA; channel self-attention; CSA; insulators.
DOI: 10.1504/IJICT.2025.144015
International Journal of Information and Communication Technology, 2025 Vol.26 No.1, pp.1 - 22
Received: 11 Oct 2024
Accepted: 06 Nov 2024
Published online: 20 Jan 2025 *