Due to the varying reflectivity of LiDAR scanning small targets on insulators at different positions, the point cloud granularity exhibits irregularity and disorder. Using conventional linear learning methods for point cloud segmentation of insulator small targets results in unreasonable segmenta-tion granularity and significant segmentation errors. Therefore, this paper proposes a point cloud subdivision optimization model for power tower insulators based on point network+LSTM. A classification network for insulator positioning timestamp data was constructed using a point network, and each point cloud feature was extracted using MLP to address the impact of irregu-larity and disorder in the point cloud on segmentation rationality. Utilizing the symmetric function MaxPooling network for secondary extraction of point cloud features, achieving three-dimensional coordinate displacement invariance with increased point cloud data volume. Remove interference from point cloud data outside the range through direct filtering and cloth filtering algorithms. Uti-lizing the nonlinear learning ability of LSTM network to improve the data dependency problem of RNN network. By iteratively training granularity through granularity multiplication, the rationality of point cloud data segmentation granularity can be improved. Establish a point cloud data seg-mentation model for power tower insulators based on LSTM network, and introduce a loss function to optimize it, in order to reduce the segmentation error of insulator small targets. The experimental results show that after the application of this method, it still maintains high segmentation confi-dence when the amount of point cloud data of power tower insulators increases. The fine-grained rationality of point cloud data segmentation is strong, and the average absolute error of segmen-tation is small, less than 0.1%, and it can preserve the details of the original data. This method can improve the reliability of point cloud subdivision of power tower insulators, with good subdivision effect, and provide reference for the classification and classification of insulator states.
