Research On Product Recommendation Method Based On Deep Learning In BIM Environment

The digitalization of the construction industry has been included in the future development planning of the Chinese government,and Building Information Modeling(BIM)has become the most promising technology in the future construction industry.In recent years,BIM technology has become more and more mature in architectural design,construction and management,making researchers constantly explore new application scenarios of BIM.At present,extending BIM technology to the field of interior design has become the key to barrier-free communication between customers and designers.The effects of BIM interior design are presented to customers in the form of 3D visualization,so that customers can intuitively feel the design results and directly participate in the design.Through a large number of furniture models contained in BIM,customers can freely place and design in 3d space,realizing immersive free design.However,how to purchase the interior items in the design model has become a difficult problem for both designers and customers.Although e-commerce technology can be seen everywhere in daily life,it still cannot effectively connect the information islands between virtual models and real goods.This problem obviously affects the promotion of BIM interior design and has become an unexplored area in the research of BIM technology.This paper aims at the problem that the dimensional difference between the BIM 3D model and the commodity image in e-commerce makes it impossible to connect the physical commodity through the 3D model,and the visual domain gap between the 3D model image and the commodity image makes it difficult to accurately find the corresponding commodity.This paper proposes a product recommendation method based on deep learning in the BIM environment.The main research work and research results are as follows:(1)A BIM optimal feature perspective extraction method is proposed for BIM model dimensionality reduction and evaluation of optimal feature perspectives.In this paper,by studying the existing model feature representation and product recommendation methods,it is found that the current method has little research on cross-dimensional product recommendation and the existing multi-view model feature representation methods do not have a standard quantitative evaluation method to select the best feature view map..To this end,this paper proposes an optimal feature perspective extraction method for BIM to solve such problems.This method firstly performs dimensionality reduction operation for the BIM model to achieve the same dimension as the product image,which is achieved by extracting different perspective views of the BIM model.In order to fully consider the different perspective characteristics of the 3D model and the computational difficulty of the computer,the perspective is fixed at the midpoint of the plane polygon of the football to extract multiple perspective maps as an alternative.Then,the candidate viewpoint maps are evaluated using deep learning techniques combined with the proposed evaluation function,and the viewpoint maps that best represent the model’s features are extracted from them.The proposed method can transform the BIM 3D model into a 2D image representing the model,and achieve the purpose of improving the accuracy of product recommendation in one dimension with the product image.During the experiment,this paper uses the proposed method to verify the product recommendation model based on deep learning technology.The experimental results show that the method can effectively improve the performance of product recommendation.(2)A cross-domain commodity recommendation method for BIM models in the BIM environment is proposed.Most of the existing cross-domain retrieval methods transfer image features from different visual domains to the same space for similarity comparison through feature transfer,so as to bridge the domain gap between different visual domains.However,the existing methods have high requirements on the number of training samples,and it is difficult to obtain enough training samples in the BIM environment.To this end,this paper proposes a BIM model cross-domain product recommendation method to complete the cross-domain recommendation between BIM images and product images.The method firstly transfers the BIM image to the commodity image domain by means of image domain transfer to achieve the purpose of bridging the visual domain difference.The method is implemented by a Cycle GAN generative model that does not need to map samples one-to-one.On the other hand,as the generative model generates generators with different weights during training,it is difficult to determine when the most suitable generator will be generated.To do this,find the best generator for generating images by adding a detection module after the generative model.The experimental results show that the proposed BIM model cross-domain product recommendation method solves the problem of visual domain gap and effectively improves the accuracy of product recommendation.This paper proposes an optimal feature view extraction method for BIM model,which solves the dimension difference between BIM model and product image,and improves the accuracy of product recommendation.The proposed BIM model cross-domain product recommendation method solves the visual domain difference between BIM and product image,and further improves the accuracy of product recommendation.This paper plays an important role in promoting the further promotion of BIM technology in the field of construction,and fills a blank for the research of BIM technology in the field of commodity recommendation.