A CNN model to classify traditional ethnic clothing style garments

Abstract

Theclassificationoftraditionalgarmentsisintheearlystagesofexplorationwithintheacademic study of fashion technology, with relatively few studies dedicated to this specific area. Despite the growing interest in applying computer vision techniques to the fashion industry, current research predominantly focuses on modern fashion styles, leaving a need for more attention to culturally significant traditional clothing. This research aims to address this need by utilizingConvolutionalNeuralNetworks(CNNs)toclassifyfemininetraditionalethnicclothing styles, specifically garments from four nationalities: German (dirndl), Indian (saree), Japanese (kimono),andSpanish(flamenco). Byleveragingdeeplearningtechniques,particularlyCNNs, we seek to enhance the accuracy and efficiency of fashion classification systems for traditional garments. Our study involves the construction of a diverse dataset with four nationality labels withatotalof5,879imagesandthedevelopingofanovelclassifiermodel,precededbyacomprehensive literature review that examines existing classification techniques and the contexts in which various ethnic clothing types have been studied academically. This review underscores the limited but growing work focusing on traditional garments. We empirically evaluate the performance of our model against established models, comparing its effectiveness in recognizing and categorizing traditional clothing styles from different cultures.As result, we achieved an accuracy rate of 94%, and loss 21% demonstrating the model’s effectiveness in classifying traditional ethnic garments. Additionaly, observing from the confusion matrix, the model classifiedcorrectlyfromthetestdatasetgerman(95.88%),indian(97.54%),japanese(92.59%)and spanish (89.22%). These findings are expected to contribute significantly to computer vision and fashion technology, offering new insights and practical applications for the fashion industry. By focusing on traditional garments, our work not only enhances the technical capabilities of fashion classification systems but also promotes a deeper appreciation and understanding of diverse cultural heritages through technology.