Research on Recognition of Bamboo and Silk Characters in Mawangdui Han Tomb Based on Deep Learning
By constructing a model for text recognition of the Mawangdui bamboo slips,it can be used to recognize the text of the bamboo slips unearthed at the same time,improving the efficiency of text recognition for researchers of bamboo slips.In addition,it also provides research ideas and technical routes for building ancient handwritten character recognition.The original dataset was enhanced by using the BAGAN-GP generative adversarial network combined with the traditional data augmentation method,and then Five image classification networks were used to carry out a comparative experiment on Mawangdui handwriting text recognition.The recognition accuracy of the model trained on the enriched balanced dataset is 98.589%,which is 9.181%higher than that of the model trained on the original dataset.In the comparison experiment,EfficientNet V2 small performed the best,with an accuracy rate of 99.048%.The combination of BGAN-GP generative model and traditional data augmentation method can be well applied to the expansion of Mawangdui simple handwritten text dataset.The enriched balanced dataset can achieve high recognition accuracy on different image classification networks.Combined with the transfer learning method and the introduction of pre-training weights,the training of the model can converge faster,and the accuracy can also be improved.
deep learningBAGAN-GPResNetdata augmentationtransfer learning
万方数据
维普
