激光点云深度学习的输电杆塔部件分割研究

Part segmentation of transmission tower based on laser point cloud deep learning

针对无人机电力巡检过程中对输电杆塔精细化巡检时,现有的杆塔部件点云提取精度不高,难以满足无人机自主巡检航线规划以及数字化信息管理等问题,该文使用基于点云数据的深度学习方法进行输电杆塔部件分割,并提出PCTTS模型,实现点云输电杆塔各部件的精准提取。首先,对原始输电走廊点云数据进行数据预处理,将数据质量较完整的杆塔作为样本;其次,对输电杆塔点云数据进行八叉树采样,抽稀点云的同时尽可能保留更多局部特征;最后,将样本输入深度学习模型,在保证平移、旋转不变性的同时,结合多尺度特征提取策略与Offset注意力机制,完成特征提取与传播,实现点云输电杆塔的部件分割。经实验验证,该文提出的方法在自建的部件分割数据集上mIOU达到94.1%,分割精度优于PointNet、PointNet++、DGCNN、Point Transformer、PointMLP等点云分割的方法。

For the UAV electric power inspection process of transmission tower refined inspection,the existing tower parts point cloud extraction accuracy is low,and it is difficult to meet the UAV autonomous inspection route planning as well as digital information management.In this paper,segmentation of transmission tower components using deep learning method based on point cloud data and proposed PCTTS model to realize accurate extraction of each component of point cloud transmission tower.Firstly,the raw transmission corridor point cloud data were subjected to data preprocessing,and the towers with complete data quality were used as samples.Secondly,the transmission tower point cloud data is sampled from an octree,and the point cloud is thinned while retaining as many local features as possible.Finally,the samples are fed into the deep learning model,which combines multi-scale feature extraction strategy and Offset-attention mechanism while ensuring translation and rotation invariance to complete feature extraction and propagation,and realize the segmentation of the parts of the point cloud transmission towers.The experimental results show that the model achieves mIOU of 94.1%on the self-built part segmentation dataset,and the segmentation accuracy outperforms that of the methods for point cloud segmentation,such as PointNet,PointNet++,DGCNN,Point Transformer,and PointMLP.