一种基于DensePoint的牙颌模型语义分割方法  

A semantic segmentation method for the dental model based on DensePoint

作　　者：马天[1] 翟洁晨 杨逸舟 杨嘉怡 刘佳[2] MA Tian;ZHAI Jiechen;YANG Yizhou;YANG Jiayi;LIU Jia(College of Computer Science and Technology,Xi’an University of Science and Technology,Xi’an 710054,China;College of Science,Xi’an University of Science and Technology,Xi’an 710054,China)

机构地区：[1]西安科技大学计算机科学与技术学院,西安710054 [2]西安科技大学理学院,西安710054

出　　处：《重庆理工大学学报（自然科学）》2022年第11期118-126,共9页Journal of Chongqing University of Technology：Natural Science

基　　金：国家自然科学基金项目(62101432);陕西省自然科学基础研究计划项目(2022JM-508);西安市科技局计划项目(21RGSF0017)。

摘　　要：牙颌模型分割是虚拟正畸系统的关键环节,针对传统的分割方法人工干预较多且交互操作复杂的问题,提出一种基于DensePoint的端到端牙颌模型语义分割方法。对牙颌模型的三维点云数据进行采样、标注和扩增处理,以满足分割网络训练要求;结合U-Net设计基于DensePoint的牙颌模型语义分割网络,同时面向点云数据集对网络下采样过程中的局部空间参数进行适应性优化,以确保网络能够提取到有效的局部特征;在Python环境和Pytorch框架中进行实现,并在增强的点云数据集上进行对比实验。结果表明:该方法的分割效果更优,准确率约90%,并且在相邻的牙齿边界和畸形的牙齿模型上具有更好的鲁棒性,较好地满足了虚拟正畸系统智能化的需求。The segmentation of the dental model is the key to the virtual orthodontic system.Aiming at the problem that traditional segmentation methods have a number of manual interventions and complex interaction operation,this paper proposes an end-to-end semantic segmentation method of the dental model based on DensePoint.Firstly,the 3D point cloud data of the dental model are sampled,annotated and amplified to meet the training requirements of the segmentation network.Then,this paper combines U-Net to design a DensePoint-based semantic segmentation network for the dental model,and conducts an adaptive optimization for the local spatial parameters of the point cloud dataset in the network sampling process to ensure that the network can extract effective local features.Finally,it is implemented in Python environment and Pytorch framework,and the comparison experiment is carried out on the enhanced point cloud dataset.The results show that,with an accuracy of about 90%,the segmentation effect of the proposed method is better and it has better robustness on adjacent tooth boundaries and malformed tooth models,which can better meet the requirements of the intelligent virtual orthodontic system.

关 键 词：牙颌模型分割 三维点云 虚拟正畸 分割网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]