基于深度学习的小样本零件模型分类  

Classification of small sample part models based on deep learning

作　　者：张维 岳洋 苗耀锋 ZHANG Wei;YUE Yang;MIAO Yao-feng(Engineering Institute,Xi’an International University,Xi’an 710077,China)

机构地区：[1]西安外事学院工学院,西安710077

出　　处：《信息技术》2024年第7期71-75,83,共6页Information Technology

基　　金：陕西省教育科学规划课题(SGH23Y2897);陕西省教育厅科学研究计划项目(23JK0630);陕西省自然科学基础研究计划-面上项目(2021JM-528)。

摘　　要：针对装备制造业中的小样本的机械零件自动化分类精度低的问题,提出一种基于辅助分类生成对抗网络(ACGAN)的机械零件分类检测方法。首先,基于PCL点云技术,获取具有代表性的阀座等10种类别的零件在固定视角下的二维视图信息;其次,采用小样本数据分类的ACGAN算法,以点云技术获取每个零件的42张二维视图为算法模型的数据集,完成算法模型的设计和训练。最终结果表明,训练集和测试集的准确率分别可达到99.80%和99.67%。因此,该算法可以高精度地实现对小样本的机械零件的自动化分类。To solve the problem of low automatic classification accuracy of mechanical parts with small samples in equipment manufacturing industry,a classification detection method of mechanical parts based on ACGAN is proposed.Firstly,based on PCL point cloud technology,the two-dimensional view information of 10 kinds of representative parts such as valve seats under a fixed visual angle is obtained.Secondly,the ACGAN algorithm based on small sample data classification is selected,and 42 two-dimensional views of each part obtained by point cloud technology are used as the data set of the algorithm model to complete the design and training of the algorithm model.The experiment results show that the training set accuracy of the two-dimensional view obtained by PCL point cloud can reach 99.80%,and the test set accuracy can reach 99.67%.Therefore,this algorithm can realize the automatic classification of small sample mechanical parts with high accuracy.

关 键 词：计算机应用技术 小样本 点云 深度学习 二维视图 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]