Research On 3D Model Recognition And Retrieval Based On Deep Learning

With the continuous development of the network and information technology,a large number of 3D models have been generated,and its recognition and retrieval technology have gradually attracted the attention of academic researchers and many industry research departments.At present,deep learning algorithms have made great progress in the field of the computer vision,especially in the fields of graphic image recognition,natural language processing and speech recognition.This article draws on the recognition and retrieval methods of other multimedia information based on deep learning,and considers applying deep learning algorithms to the classification and retrieval of 3D models.The main work is as follows:The shape features of the 3D model are extracted by the projection method.First,the 3D model is translated,zoomed and rotated.Then,the three view projection view of the 3D model is extracted,and then the three view projection view of the model is processed by means of mean and prewhitening,dividing the block and so on,which can be transformed into the form of the network input in the deep learning network.An improved sparse noise reduction self-coding network 3D model recognition method is proposed.The method uses a sparse noise reduction self-coding model combined with an improved unsupervised and restricted quasi-Newton calculation method to construct a deep neural network,thus processing the 3D model information deeply and then classifying and recognizing the 3d model.A three-dimensional model recognition method based on topological sparse coding is proposed.This method uses an improved unsupervised feature learning algorithm of topological sparse coding model to construct a deep neural network,and uses greedy algorithm to iteratively optimize the cost function JsAof the network model,learning),(the weight matrix A and outputting the visual feature A,which effectively improves the shortcomings that the common sparse self coding model extracting feature time is long.A weighted optimized deep convolution network is proposed,and the convolution network of three side projection images is calculated.The network is applied to the recognition and retrieval of 3D model.Experiments show the effectiveness of the method.