Application of Convolutional Neural Network in B-Rep Models Classification

Aiming at the problems of expensive calculation and complex feature extraction of existing 3D models classification methods, this paper proposes a classification method based on convolutional neural network (CNN). This paper uses multi-view to represent 3D models, views contain information from multiple aspects of the model, and they have certain links. Constructing a convolutional neural network model, uses the features extracted from the multiple layers as a strongest descriptor. The classifier selects Softmax regression to solve the multiple classification experiments. The experimental results show that in 3D models classification CNN+Softmax had a higher accuracy rate compared to the traditional 3D models classification methods, whose accuracy rate is 86%.