Deep learning-based pattern classification for embroidery in Asia: evidence from Kazakh and Kyrgyz patterns

Asian ethnic embroidery embodies the culture, history and aesthetics of ethnic groups through animal, plant, and geometry pattern. However, due to the abstract nature and cultural exchange, these patterns often share overlapping elements, making accurate classification challenging. This can lead to fundamental misunderstandings toward the culture behind, and hinder the preservation and inheritance of Asian embroidery heritage. Thus, this study conducts a case study on Kazakh embroidery patterns for model selection and evaluates several deep learning networks, including ResNet variants with and without the Convolutional Block Attention Module (CBAM). The selected best-performing network is subsequently assessed on Kyrgyz embroidery patterns for validation. The results show that ResNet50-CBAM exhibits best performance across fold, pattern and network performance. This study provides a methodological reference for ethnic embroidery studies and offers a potential tool for the identification, documentation, application, and preservation of ethnic embroidery patterns in Asia.