A Point Cloud Recognition Method for Substation Equipment Based on Improved Point Transformer

A point cloud recognition method for substation equipment based on improved Point Transformer is proposed to address the issue of low accuracy in point cloud segmentation during multi-target recognition. First, a new dot product and double random inter-point attention mechanism embedding are created in the convolutional projection layer of the Point Transformer model. The dynamic adaptive graph convolution processing of local neighbourhoods is utilised to continuously increase global feature attributes in the data feature vector in order to improve the accuracy of point cloud recognition. Then, unmanned aerial vehicles and inspection robots are used to collect point cloud information of substation equipment, which is input into an improved Point Transformer model for analysis to obtain the three-dimensional point cloud recognition results of the equipment. Finally, the point cloud recognition results are input into a two-dimensional space for fast Euclidean clustering to obtain the three-dimensional point cloud instantiation recognition results of substation equipment. The three-dimensional point cloud data collected from a 110 kV substation is selected for experimental testing of the proposed method. The result shows that it has good recognition performance for all devices, with an average accuracy of 94.30% and a time consumption of 3.38 s, which is superior to other comparative methods.