| Ethnic culture plays an indispensable role in the development of our country.Ethnic cultural elements are important units that constitute ethnic culture,mainly including elements of ethnic culture such as clothing,food,housing and transportation,etc.However,some elements of ethnic culture in network are gradually declining or even disappearing with the progress of technology.So the efficient extraction and protection of ethnic cultural images in cyberspace is one of the most important works in ethnic culture research undoubtedly.This paper focus on the key technology of extracting ethnic cultural elements in images.The purpose of this paper is to explore the effective ways of extracting and protecting ethnic culture elements in network images,and to provide technical support for the safe development of Chinese ethnic culture in cyberspace.The research content includes the following three aspects.(1)Research on the extraction and detection method of ethnic costume elements based on attention mechanism.Firstly,this paper summarizes the local feature attributes of ethnic costumes,divides the costumes into six parts according to the human body structure.This paper defines the fine-grained semantic attribute labels of each part,and designs the ethnic costume element extraction and detection method with the characteristics of regional object detection.Secondly,Faster R-CNN is improved based on SENet attention mechanism.SE-Block is added after the backbone network Res Net50 to make the network adaptively weighted channels.This method is to make Faster R-CNN emphasize important features and weaken useless features to achieve the purpose of improving the accuracy in ethnic costume region detection.Local marking is performed according to the labels to construct the dataset MZdata1 containing Mongolian,Korean and Manchu costumes in Jilin Province.Finally,the effectiveness of the proposed method is verified by experiments.(2)Research on the extraction and classification method of ethnic costume elements based on global detection and local segmentation.Firstly,this paper designs the element extraction and classification model for parallel processing.The image background information is removed by U2-Net.The image enters the Faster R-CNN based on SENet and VGG16 based on dilated convolution to obtain the local features and global features respectively.And then the prediction category X and Y are obtained.If X is equal to Y,the image category is output,otherwise it is discarded.Next,MZdata2 is obtained by adding global marking to MZdata1.U2-Net avoids the interference of background information on feature extraction.VGG16 based on dilated convolution makes the extraction of ethnic costume elements increase the receptive field without changing the feature layer.The model uses parallel processing to improve the accuracy of classification.Finally,the feasibility of the proposed method is verified by experiments.(3)Research on ethnic culture element extraction and automatic recognition system based on convolutional neural network.Firstly,the element extraction and automatic recognition system is designed.Improving Dense Net201 based on the dropout strategy to solve the overfitting problem of ethnic culture elements due to the small amount of data.The dataset MZdata3 containing cultural elements of Mongolian,Korean and Manchu clothing,food,housing,transportation and musical instrument in Jilin Province is constructed and globally marked.Secondly,the parameters that make the model optimal are selected by experimental comparison of the weight initialization.The effectiveness of the proposed method is verified by experiments.Finally,a visualization system is designed using Py Qt.The paper also compares and analyzes the changes of the four ethnic cultural elements in one year. |
