An Intelligent Point Cloud Recognition Method for Substation Equipment Based on Multiscale Self-Attention

The semantic recognition of point cloud is an important aspect of point cloud applications, it is crucial to study the intelligent point cloud recognition method for substation equipment to replace manual processing. Based on previous related work, this paper deals with the construction of an intelligent point cloud recognition network for substation equipment. In this network, a point cloud to tensor (PC2T) module is proposed to achieve the goal of using a 2D neural network to process 3D point cloud; a multi-scale self-attention (MSSA) module is introduced to optimize the global feature extraction of point cloud and enhance the accuracy of point cloud recognition. In addition, this paper proposes two point cloud data argumentation methods to assist in network training. The experimental results show that the constructed point cloud recognition method has excellent recognition accuracy and point cloud quality robustness, and that the proposed data argumentation methods are effective. The research results can provide a reference for the digital transformation of substations.