基于卷积自编码网络的杆塔点云数据自动分类方法

在输电线路三维可视化自动建模场景中如何实现对杆塔的三维点云数据进行快速准确的自动化分类是一个关键问题,在本文中我们提出了一种基于卷积自编码神经网络CAE的杆塔三维点云数据自动分类算法。首先,我们通过投影计算得到杆塔点云的旋转角度并使用旋转矩阵将杆塔点云摆正,然后进行正面侧面投影获取到杆塔点云的图像;第二,使用收集到的杆塔点云图像组成训练数据集,对卷积自编码网络进行训练之后提取出自编码网络的编码部分用于对图像进行特征提取;第三,使用自编码器对输入的杆塔点云图进行特征抽取,将提取的图像特征向量输入EM进行自动分类。实验结果表明我们所提出的杆塔点云自动分类算法能够快速准确实现对点云数据的自动化分类。

Classification method of tower point cloud data based on convolutional auto-encode

How to realize the fast and accurate automatic classification of the three-dimensional point cloud data of tower in the three-dimensional visual automatic modeling scene of transmission line is a key problem.In this paper,we propose an automatic classification algorithm of the three-dimensional point cloud data of tower based on convolutional auto-encode(CAE).First of all,we obtain the rotation angle of tower point cloud by projection calculation,and use rotation matrix to straighten the tower point cloud and carry out front side projection to obtain the side view image of tower point cloud;second,use the collected image of tower point cloud to form the training data set,and then train the convolutional auto-encode.The coding part of the CAE is used to extract the features of the image;thirdly,the CAE is used to extract the features of the input image of tower point cloud and the extracted image feature vectors are input into EM for automatic classification.