The Script Recognition Technology Of The Bamboo Slips In The Warring States Period Based On Deep Metric Learning

Chinese character has a long history,and there has also been a long-term and in-depth studies on linguistics and historical documentation.With the rapid development of computer artificial intelligence technology,characterrecognition technology has been integrated into every field of our lives,and the use of information processing technology for ancient character research has also attracted more and more scholar’s attention.At present,preliminary research has been made on the intelligent recognition of Oracle inscriptions,but the cross-over study between the study of characters in the Warring States Period and the computer character recognition is still relatively rare.Therefore,in this academic paper,the few shot learning algorithm based on the siamese network is used to identify the character of the Warring States Period.Main tasks as follows:(1)Based on the fact that there is no suitable data set for the identification of the character in the Warring States period,this paper proposes and collates the Bambooslips data set.For the documents of the Warring States Period,which were initially sorted out by Anhui University,each bamboo slip was taken out separately under the premise of maintaining high resolution,and stored in correspondence with the Chinese character file.For the currently unavailable character,it is realized by "making characters".Use the labeling software to cut and save the individual character in each bamboo slip,and save the pictures of the same character in a folder to form the Bambooslips data set.It is used for the recognition of the abbreviated character of the Warring States Period.(2)Considering that the Bambooslips data set belongs to the category of small samples,this paper proposes a few-shot character recognition algorithm based on the siamese network for bamboo slips.The overall recognition accuracy of the algorithm on the Bambooslips data set and the recognition accuracy of individual character are output in sequence.And introduce data enhancement transformation into the original small data to increase the capacity of the data set.On this basis,the original data and the enhanced data are trained and recognized separately,and the recognition accuracy is compared.On the Bambooslips data set,the recognition effect of the fewshot recongnition algorithm based on the siamese network of the bamboo slips is superior to that of the other deep metric learning algorithms.(3)In order to make the algorithm have practical application,this paper designs a character recognition software platform based on Pyqt.It includes the functions of importing pictures,zooming in and out of pictures,labeling text,cutting and saving a single character,identifying a single character and outputting the corresponding meanings of bamboo slips.The software can input bamboo pictures to be recognized,obtain a single character picture,then output the corresponding Chinese characters in sequence by algorithm recognition.