基于卷积神经网络的非等效点云分割方法

Non-equivalent Point Cloud Segmentation Method Based on Convolutional Neural Network

针对点云分割中的非等效性,提出基于CNN(convolutional neural network)的点云分割神经网络NEPN(non-equivalent point network),在设计网络随机抽样层与MaxPooling层解决点云数量与顺序多变的基础上,引入经惩罚函数作用后的距离矩阵对各点分类误差进行加权,优化模型训练损失计算方法,强化分割面邻近点的误差反馈,实现点云区域分割。试验结果表明,该方法分割精度优于PointNet方法,可有效解决非等效点云分割问题。

Aiming at the non-equivalence in point cloud segmentation,NEPN(non-equivalent point network),apoint cloud segmentation neural network based on CNN(convolutional neural network)was proposed.On the basis of designing the network random sampling layer and MaxPooling layer to solve the problem of the number and order of point clouds,the distance matrix after the penalty function was introduced to weight the classification error of each point.The model training loss calculation was optimized,and the error feedback of the adjacent points of segmentation surface was strengthened to realize the point cloud region segmentation.The experimental results show that the segmentation accuracy of this method is better than that of PointNet method,which can effectively solve the problem of non-equivalent point cloud segmentation.