Preliminary Research On The Identification Of Typical Doors And Windows Of Shanxi Traditional Dwellings Based On Improved YOLOv2

Shanxi traditional dwellings is a precious heritage left by history,which is not only the true portrayal of history and culture,but also the concrete manifestation of people’s spiritual pursuit at that time.However,the destruction of the cultural heritage of traditional dwellings in Shanxi Province is worrying.Under the current background of big data era,a new residential protection measures research is imminent.The 21st century is an era of information explosion,and the emerging image recognition technology is widely used in many fields for its rapid and accurate information extraction and recognition.Based on the status,this paper proposes the research idea of combining the image recognition technology with the information protection work of traditional residence,which is bound to create a new means of heritage protection and comprehensively improve the protection of historical buildings.Doors and windows is a part of traditional dwelling components which is easy to be identified and have the largest number of house components.This paper firstly studies the relevant theories of typical doors and windows in Shanxi traditional dwellings,including the types,functions and carvings of doors and windows.Secondly,the image recognition technology is explained and expounded.Then,on the basis of a large number of field research on traditional dwellings in Shanxi Province,the first-hand data were collected.In view of the current situation that relevant research directions are lacking of the image recognition of traditional dwellings’ doors and windows,this paper proposes a typical image recognition algorithm of traditional dwellings’ doors and windows in Shanxi Province based on the improved YOLOv2 network model which is called YOLOv2-TDDWNet.The improvement measures mainly include:·Adding network training process visualization part;·Adding BN network layer to the original network model;·Adding convolution layer to improve the accuracy of recognition;·Designing graphical user interface(GUI)to improve network identification and detection.In order to verify the effectiveness of the detection method,the paper then compare this netework with other improved models.The experiment shows that the improved YOLOv2-TDDWNet model achieves an average accuracy of nearly 80%in the detection of six types of traditional dwellings’ doors and windows under single target and multiple targets,and it has an excellent recognition effect.This paper explores a new research method for the protection and inheritance of traditional dwelling doors and windows by means of interdisciplinary integration.The research content is no longer limited to the static study of single doors and windows of traditional dwellings,but to provide convenient means for the data research of doors and windows from the perspective of identification with the help of Deep Learning.The research method is no longer limited to the survey,mapping and picture description based on architecture,but is gradually moving towards the data-based exploration.