3D Model Classification Based On Shape Feature And CNN-LSTM Network

With wide application and continuous development of 3D modeling technology,the amount of data and the complexity of 3D model are increasing day by day.As important part of digital geometry,3D model classification has become important research content in the field of computer graphics and computer vision.This paper studies a 3D model classification method based on shape features and CNN-LSTM network.The overall shape feature and local shape feature of3 D model are used to express geometric shape of 3D model.CNN network and LSTM network are combined into a 3D model classification model based on CNN-LSTM network.This article mainly studies the following:1.The research background and research status of 3D model classification are studied.Methods of extracting features from 3D models and 3D model classification methods based on deep learning are introduced.2.Methods of extracting geometric features from 3D models are studied,and the overall shape features D1,D2,D3,A3 and local shape features SDF are used to express the geometry of 3D model.3.3D model classification algorithm based on convolutional neural network is studied.CNN consists of convolutional layer,pooling layer and fully connected layer.The network model is optimized by shape features of the training data,shape features of test data are input into the network model for classification experiments.4.The characteristics of cell structure in LSTM network are studied.LSTM network uses cell structure to learn the sequence information in the feature sequence.Based on shape features of 3D model,the 3D model classification based on LSTM network is constructed.5.3D model classification based on shape features and CNN-LSTM network is studied.CNN and LSTM network are combined,in which LSTM layer is connected after CNN layer.The purpose is to utilize the advantages of two networks.Shape features in training data are used as the input of the CNN-LSTM network,their categories are used as the output of the network model.Parameters in CNN-LSTM network model are optimized by back propagation.Shape features of test data are input into the optimized CNN-LSTM network model for classification experiments.