Research On Educational Resources Retrieval Of Minority Costume Image Based On Convolution Neural Network

With the rapid growth of digital media resources in the Internet age,the image of minority costume has become the main presentation of the digitalization of minority costume culture.As one of the vital contents of minority culture,minority costume images not only provide extremely important educational resources for minority education informatization,but also greatly promote the inheritance and protection of minority costume culture.Accordingly,how to use modern information technology to deal with this essential ethnic educational resources and protect it has become a hot issue.Due to the diversity of styles,colorful colors and rich patterns,the visual features of ethnic costumes are difficult to be accurately described and extracted.Therefore,using the traditional image retrieval technology based on low layer features to retrieve the educational resources of national costume image can not get satisfactory results.In recent years,deep learning has been studied and successfully applied in the field of image.In view of the prominent advantage of convolution neural network in automatic acquisition of high level semantic features,this paper proposed a retrieval method based on convolution neural network for minority costume image education resources,which provides a technical reference for the related research in this field,and plays a great role in promoting the protection and inheritance of ethnic cultures.The main research substance are as follows:(1)The acquisition of Yunnan minority costume pictures data and the construction of the resource library.The data of costume images resource library are taken from some ethnic minority areas in Yunnan via our research team.At present,the database mainly contains the Hani,Va,Yi minorities' clothing images of three ethnic minorities.The construction of resource library provides data support for the algorithm which researched in my paper.At the same time,it also provides data basis for the researchers in the related fields.(2)In-depth study of traditional image retrieval technology and image retrieval technology based on depth learning.Through depth analysis image retrieval algorithm based on low-level feature extraction,the disadvantages of traditional retrieval technology were pointed out which are the low accuracy and high complex calculation.The structure and algorithms of neural network were studied in-depth,this paper proposed a retrieval method based on convolution neural network for minority costume image education resources.Using the OpenCV image library and the Caffe neural network framework as the development environment,combined with the minority dress image education resource.Firstly,Preprocessing training samples by filtering denoising technique.Then,using three network structure models of AlexNet,LeNet and CaffeNet to train minority costumes image samples respectively,and finally got the neural network training model of minority costume images.(3)On the minority costume education resource base constructed in this paper.By compared with classical algorithms,HSV,LBP,three order Hu invariant moments and LIRE,it is proved that convolution neural network has higher retrieval accuracy and faster retrieval to the Minority costume image retrieval.(4)Under the platform of Windows10,taking C++ as the main development language,and based on the image retrieval algorithm of convolution neural network proposed in this paper,I have designed and implemented the minority costume image education resource retrieval system.It mainly includes the reading of data,the processing of image,the recognition of national costume,and the storage of data.