Research On Intelligent Indoor Scene Matching Technology Based On Deep Learning

At present,VR technology has been widely used in the field of interior design.The interior design was presented to the clients by designers via VR panoramic roaming.A large number of VR panorama solutions have emerged on websites for users to choose.With the traditional text-based retrieval method,it is difficult to accurately find the panoramic solution required by the owner in a large number of solutions.A VR panoramic scheme matching method based on image intelligent retrieval will be studied.By using deep learning technology,similar schemes can be quickly and accurately matched in a large number of VR panoramic schemes through an indoor picture.The main content of the paper is as follows:(1)The feature of VR panorama scheme is analyzed.And the retrieval of VR panorama scheme is transformed into the image content matching.As artificial feature design is difficult to extract the precise and complete features of the image,the relevant theoretical knowledge of deep learning is explained.And the advantages of convolutional neural networks for image feature extraction have been analyzed,which provides theoretical basis.(2)A method of automatic extraction of image features and labels based on convolutional neural networks has been studied.The convolutional neural network model was designed,which is optimized by dropout.ELU activation function was introduced to solve the problem that Re Lu activation function has no negative value.Experiments show that this method speeds up training and improves classification accuracy.Improved softmax classifier is used to solve the problem of softmax calculation overflow.(3)A fast retrieval algorithm that combines text and content through pictures was studied to solve the problem of retrieving VR panorama scheme and the problem of semantic gap based on content retrieval.This algorithm combines the advantages of text retrieval and content retrieval.It can solve the problem that the VR panoramic scheme is difficult to describe in text and similar in semantics.The VR panorama solution database is designed to convert the panorama solution into pictures.Convolutional neural networks are used to automatically extract image tags and features.PCA technology is used to speed up the algorithm.(4)A retrieval system is designed and implemented based on the method of this article.The system enters a picture and matches the VR panorama scheme in the database through the method of this article.The system includes the picture input module,CNN network model module,image classification module,PCA module,retrieval algorithm module,database module and scheme output module.With interface-oriented programming ideas,each module is designed independently to achieve module decoupling.