关键词: 稀疏降噪自编码, 三维模型识别, softmax分类器

Abstract: Aiming at the problem of low accuracy of 3D model feature recognition in massive data mining, we proposed an improved sparse denoising autoencoder neural network model. Firstly, based on the improved sparse denoising autoencoder method, we constructed a deep neural network model, then used the unsupervisedpretraining method and restricted quasi Newton method to train the autoencoder neural network. Finally, the softmax regression and the obtained features were used to train the final classifier. The results show that the method has well robustness to the feature information of the three\|dimensional model with noise. Compared with the stack of autoencoder neural network and the selflearning neural network, the method has better recognition rate.

Key words: sparse denoising autoencoder, 3D model recognition, softmax classifier