Research On The Automatic Measurement Method Of Ancient Hanfu Size

The study on the structure and shape of ancient Han clothing is an important part of the study on traditional clothing.In order to avoid damage to cultural relics caused by weathering and other problems,most of the unearthed clothes will be kept in the warehouse after measurement and photo taking,and rarely displayed.Moreover,due to the limited data given in the excavation report,subsequent researchers cannot carry out size measurement on demand,which is not good for the inheritance and development of traditional clothing culture.To solve above problems,this paper proposes using convolution neural network to design a set of ancient hanfu size automatic measurement scheme,this scheme can under the condition of the sample data are scarce,rapidly and exactly by machine learning to obtain the size of the ancient hanfu data,go to the lavatory not only traditional clothing rehabilitation and research work,and the artificial measurement results review can be used in the museum.Since there are many types of hanfu and various shapes and forms,in order to ensure the accuracy of the measurement results,data sets should be made according to the clothing types of hanfu for parameter training during model training.This paper takes the most representative short-cut jackets of Hanfu as the experiment object,including Ru,Shan and Ao.Firstly,a two-stage convolutional neural network model Re-Global Net-Refinenet was built to detect the key points of Hanfu.The model was transferred and iterated by using the modern clothing data set containing 50000 jacket pictures and the Hanfu data set containing 5000 jacket pictures to improve the model’s generalization ability.Then the algorithm is used to detect and cluster the key points outputted by Re-Global NetRefinenet,and the effective information of key points of Hanfu-Fu is obtained.Finally,based on the tiled map of Hanfu given in the museum or excavation report and a real measurement size,the size data of the whole garment can be obtained through the algorithm mapping.The experimental results show that the trained convolutional neural network model Re-Global Net-Refinenet has a high degree of convergence and good training effect.The missed recognition rate of key points is 0,the single point accuracy rate is 94.05%,and the recognition accuracy rate is 81.3%.The coordinate points meet the error range.The absolute error of the garment size measured by this scheme is between 2-10 mm,and the relative error is between0.50%-5.00%,which meets the measurement error range given in the tolerance range table of garment size.This scheme provides a new idea for the restoration research of traditional clothing and the measurement of the size of cultural relics.This article consists of six chapters.The first and last chapters are the introduction and the conclusion.The middle chapters introduce "Convolutional Neural Network Basics and Optimization,Convolutional Neural Network Optimization Design,Data Set Production and Model Training,Automatic Implementation of the measurement plan process,system evaluation and comparison experiment".