Research On The Segmentation Of Ethnic Minority Costume Patterns Based On Attention Mechanism

Ethnic costume culture is a reflection of the intermingling of various cultures,including the fruits of labour,colour preferences,customs and religious beliefs of various ethnic groups,and is an essential part of traditional Chinese culture.With the rapid development of human society and the collision and intermingling of various cultures,important elements of ethnic minority costumes are constantly being lost.Therefore,segmentation of accurate ethnic minority costume patterns can effectively preserve them digitally and play an extremely important role in the analysis and interpretation of ethnic minority costume culture.Ethnic costumes are diverse,reflecting the fact that they contain unique patterns,totems,textures,colours and other important features.This paper uses the Seg Net model as the backbone network for pattern segmentation research.The segmentation effect of the model in the minority costume patterns dataset has not achieved good results,and how to improve the Seg Net model for accurate segmentation of the minority costume patterns dataset is of great significance to the promotion and development of minority costume culture.The main research elements of this paper are as follows:(1)For the problem of difficult feature extraction from target regions caused by diversified elements of ethnic minority costumes and irregular patt erns of ornamentation,a segmentation algorithm based on the SE attention mechanis m is proposed.The attention mechanism can strengthen the focus on the target regions and obtain more information about the pattern features.Firstly,the improved Seg Net model is used as the backbone network for feature extraction of ethnic minority costume patterns.Secondly,the SE attention mechanism is incorporated into the model to focus on more detailed features.Finally,the segmentation results of different network models are compared by different evaluation metrics.The experimental dataset of this paper is obtained from the Ethnic Costume Museum of Beijing Institute of Fashion,and the ethnic minority costume patterns are collected by specific image types,and the costume patterns are pre-processed relatedly for the experiments of this paper.The experimental results show that by fusing the SE attention mechanism and the improved Seg Net model,the accuracy,intersection ratio and similarity coefficient of the model are higher than those of other networks,and the segmentation results are better than those of other comparison networks.(2)A segmentation algorithm based on attentional feature fusion is proposed to address the diversity of ethnic minority ornament patterns and the difficulty of extracting target regions caused by large colour differences.Deep over-parameterized convolution is used in the Seg Net model to increase the learnable parameters and enhance the model performance,and the attention feature fusion m echanism is introduced,which can fuse the features of different layers as a way to help the model obtain more pattern information.Finally,the segmentation results are compared by different evaluation metrics and the segmentation results of ethnic minority costume patterns with large colour differences.The experimental results show that the proposed model achieves excellent results on the ethnic minority dress pattern dataset.