基于注意力机制的三维模型特征提取

Three⁃dimensional model feature extraction based on attention mechanism

目前在基于视图的三维模型检索技术中,对多视图特征提取的大多方法,关注于视图的全局特征信息而忽略了对视图局部特征信息和多视图之间的相关性的探究。针对此问题提出一种新的特征提取方法,利用深度学习中的卷积神经网络,并结合注意力机制提取特征,以提升其判别性。方法在ModelNet40上进行实验分析,将三维模型的多个视图作为输入,在网络层中加入注意力模块进行特征提取分类,结果表明,该方法在分类准确度方面优于已有的典型算法。

Currently,in view⁃based 3D model retrieval technology,most methods for multi⁃view feature extraction focus on the global feature information of the view and ignore the exploration of the correlation between the local feature information of the view and the multi⁃view.To solve this problem,a new feature extraction method is proposed,which uses the convolutional neural net⁃work in deep learning and combines the attention mechanism to extract features to improve its discriminability.The method con⁃ducts experimental analysis on ModelNet40,takes multiple views of the three⁃dimensional model as input,and adds an attention module to the network layer for feature extraction and classification.The results show that this method is superior to existing typical algorithms in terms of classification accuracy.