Fig 1.
A sample of the data set and its contents.
Fig 2.
The VGG16 network can be used as the architecture of the feature extraction network.
Table 1.
Performance of three networks combined with Siamese networks.
Fig 3.
The two images are compared against each other.
Fig 4.
The architecture of the Siamese network.
Table 2.
Experimental results of data augmentation accuracy.
Fig 5.
The Siamese network with the VGG16 network as the backbone feature extraction network predicts the evolution rules flow of Chinese characters.
Fig 6.
After feature extraction by the VGG16 network, a multi-dimensional feature is obtained and its similarity is measured.
Fig 7.
The accuracy rate of our training set.
Fig 8.
Detailed examples are given for four evolutionary rules.
Table 3.
The similarity of Chinese characters corresponds to the 4 evolution rules respectively.
Fig 9.
Accuracy between adjacent periods.