Application Of Support Vector Machine And Residual Network Model In The Recognition Of Five Ethnic Minority Costume

China is a unified multi-ethnic country,and the ethnic groups with distinct characteristics jointly constitute the Chinese national community.Therefore,ethnic culture has become the focus of research.As the carrier of the traditional culture of ethnic minorities,the ethnic costume is the precious property of the whole Chinese nation,and its intuitive presentation has triggered hot discussion and discussion.With the gradual maturity of science and information technology,how to inherit and carry forward our minority clothing culture with the help of scientific and effective methods,so that this cultural wealth can be protected has become a new topic that people in today’s society generally attach importance to.In view of this,this paper collected a lot of clothing image,based on the five types of Chinese common ethnic costumes,with the help of the relevant technology of the image processing,respectively,the traditional support vector machine(SVM)classification algorithm of machine learning and deep learning residual error of the neural network model used in ethnic costumes recognition,comparison and analysis summarizes the model experimental results.The main work and conclusions are as follows:(1)To construct the clothing image set.In view of the lack of standard and reliable data set of minority clothing image that can be downloaded at present,this paper obtained a total of 1765 clothing images in 5 categories,including Manchu,Mongolian,Miao,Yao and Zhuang,through web crawlers.The images that do not belong to the clothing category of the corresponding label and those with excessive watermarks are filtered out manually.(2)Ethnic clothing recognition based on support vector machine is realized.In this paper,clothing features are extracted by LBP texture operator and clothing features are extracted by LBP and PCA fusion.The processed eigenvector matrix is input into support vector machine classifier,and the model is trained and tested.Finally,57.55%and 63.02%recognition accuracy are achieved.(3)Realization of convolutional neural network recognition based on ResNet50.In this paper,the data expansion of the image set is realized by random clipping and other data enhancement methods.The recognition model is obtained by training the image data based on the original ResNet50 network and the optimized ResNet50 network after changing the number of output channels.Finally,the recognition of five categories of ethnic clothing is realized through the test set.Compared with the results of the study,the original ResNet50 network has a certain degree of over-fitting phenomenon in the data set.The recognition rate of the training set and test set of the optimized ResNet50 network model is relatively high,and the recognition effect is more ideal,and the accurate recognition rate of the test set reaches 94.13%.