中国科学数据（中英文网络版）2025，Vol.10Issue(2)：496-505,10.DOI:10.11922/11-6035.csd.2024.0093.zh
gZiBeads天珠图像识别数据集
A dataset of gZiBeads images for dzi beads recognition
摘要
Abstract
As a rare type of etched agate bead,dzi beads are known for their diverse symbolic meanings and intricate,varied patterns,posing certain challenges for most hobby collectors attempting to identify them in the dzi market.However,with the rapid development of deep learning technologies,traditional artificial classification methods-characterized by being time-consuming and labor-intensive,and highly dependent on experts' subjective experience-can no longer effectively meet the market's demands.Therefore,it is necessary to construct a dataset of gZiBeads images for dzi beads recognition to explore a deep learning-based automatic identification approach,which is crucial for improving the accuracy and efficiency of dzi bead identification.We firstly collected,classified and saved images of different categories of dzi beads from the Tibetan region,resulting in a dataset comprising 22 categories of dzi beads with 5,443 original images,ranging from 203 to 305 images per category,and a total data volume of approximately 693.4 MB.Secondly,to further enhance the dataset,we applied data augmentation techniques,and divided the dataset into the training set,the validation set and the test set with a ratio of 6∶1∶1.Finally,mainstream image classification methods were used to conduct multiple rounds of experimental validation on both the original and the augmented datasets,achieving a validation accuracy as high as 93.8％.The constructed dataset can effectively meet the training,validation and testing needs of current deep learning classification and recognition models,providing essential data resources for the study of dzi image classification,image segmentation,target detection and the development of the mobile platform for automatic recognition.关键词
天珠/gZiBeads数据集/深度学习/天珠图像/自动识别Key words
dzi beads/gZiBeads dataset/deep learning/dzi beads image/automatic recognition引用本文复制引用
夏建军,高定国,许松涛..gZiBeads天珠图像识别数据集[J].中国科学数据（中英文网络版）,2025,10(2):496-505,10.基金项目
国家自然科学基金(62166038). The National Natural Science Foundation of China(62166038). （62166038）
