全文下载： 
20250105 .pdf
文章编号： 1672-6987（2025）01-0120-07 DOI： 10.16351/j.1672-6987.2025.01.016
崔凤英， 李佩佩， 曹梦龙（青岛科技大学 自动化与电子工程学院，山东 青岛 266061）
摘要： 为精准捕获假肢几何细节，提高重构假肢曲面的精确度和完整性，提出了一种基于注意力机制的3D假肢重建算法。算法首先引入特征嵌入层，解决点云数据的无序性问题；然后引入残差式注意力模块，提取物体雏形特征信息同时避免网络性能出现退化；为丰富物体形状信息，采用EdgeConv模块捕获局部特征信息，并采用权重特征融合模块将其与物体的粗糙形状信息进行融合；最后将物体的精细特征和任一点特征输入到IM-Net解码器，判断该点相对于该形状的内外状态值，并通过标记立方体来提取表面。实验结果表明，与DMC、DeepSDF相比，所提算法在ModelNet40上重建任务的均方误差分别减小30%和45.7%。通过采集4位患者假肢实测数据进行仿真验证，所提算法的均方误差至少降低12.2%。因此所提出的算法捕获假肢细节特征的能力较强，重建模型精度较高，值得推广应用。
关键词： 注意力机制； 残差结构； 3D假肢； 表面重建
中图分类号： TP 391 文献标志码： A
引用格式： 崔凤英， 李佩佩， 曹梦龙.基于注意力机制的3D假肢重建算法［J］.青岛科技大学学报（自然科学版）， 2025， 46（1）： -.
CUI Fengying，LI Peipei，CAO Menglong.3D artificial limb reconstruction algorithm based on attention mechanism［J］.Journal of Qingdao University of Science and Technology（Natural Science Edition）， 2025， 46（1）： -.
3D Artificial Limb Reconstruction Algorithm Based on Attention Mechanism
CUI Fengying， LI Peipei， CAO Menglong（College of Automation and Electronic Engineering， Qingdao University of Science and Technology，Qingdao 266061， China）
Abstract： In order to accurately capture the geometric details of the prosthesis and improve the accuracy and integrity of the reconstructed prosthesis surface， this paper proposed a 3D prosthesis reconstruction algorithm based on attention mechanism. Firstly， the algorithm introduced the feature embedding layer to solve the disorder of point cloud data； After that， the method introduced the residual attention block to extract the rudimentary feature information of the object and avoid the degradation of the network performance； To enrich the characteristics of the object. The algorithm used the EdgeConv module to capture the local feature information， and the weight feature fusion module to fuse them with the rough shape information of the object. Finally， the fine feature of the object and the feature of any point are input to the IM-Net decoder to determine the value of the internal and external state of the point relative to the shape， and extracted the surface by marking the cube. Compared with DMC and DeepSDF， the mean square errors of the proposed algorithm on ModelNet40 reconstruction task reduce by 30% and 45.7%， respectively. The mean square error of the proposed algorithm reduces by at least 12.2% by collecting the measured data of four patients. Therefore， the proposed algorithm has a strong ability to capture the details of the prosthesis， and the accuracy of the reconstruction model is high， which is worthy of generalization and application.
Key words： attention mechanism； residual structure； 3D prosthesis； surface reconstruction
收稿日期： 2023-12-01
基金项目： 山东省研究生教育质量提升计划项目 （SDYJD18029）.
作者简介： 崔凤英（1972—），女，副教授.
