Chinese character writing quality evaluation method based on deep learning
Technical Field
The invention relates to a Chinese character writing quality evaluation method based on deep learning, and belongs to the field of intelligent evaluation.
Background
AI is a very common internet intelligent system, people have specific application in various scenes, such as city traffic guidance, railway transportation guidance and the like, but these are used on large-scale equipment, AI is an intelligent system which is formed through a large amount of data and runs at a high speed through learning and exercising, these systems can be intentionally circulated, regular things can be found, some things which are not regular, such as fonts, the fonts written by everyone are different, the aesthetic feeling of everyone to the fonts is different, but how to express the fonts which are all perceived as beautiful through an irregular thing can be realized, and the judgment is very difficult only by AI.
Disclosure of Invention
The technical problem to be solved by the present invention is to provide a feasible method, simple steps, which converts irregular and searchable characters into a method that can be evaluated by intelligent AI:
the invention aims to complete the technical scheme that a Chinese character writing quality evaluation method based on deep learning comprises the following steps:
the method comprises the following steps: collecting hand-written Chinese characters and establishing character library
The method comprises the steps of collecting general Chinese character handwritten word graphs with different writing qualities as much as possible, and then labeling each word graph, such as writing quality scores, character structure scores, form scores, gravity center scores, stroke scores, radical scores, similarity scores with template standard characters, structural comments, form comments, re-written comments, stroke comments, radical comments, similarity comments with templates, Chinese character writing overall comments and the like.
Step two: data extension based on generation of antagonistic neural network GAN
On the basis of the above small scale with labeled data sets, a semi-supervised deep learning model is used. Through a method of combining the recurrent neural network and the generative confrontation network GAN, the recurrent neural network learns the labeling relation and the characteristics of the data, and the generative confrontation network generates GAN reasonable data so as to expand a data set. And then, through data processing and other work, a reliable data set for model training is formed, and the problem of shortage of the data set is relieved and supplemented. Thereafter, a Deep learning convolutional neural network Deep-CNN structure is used;
step three: deep learning convolutional neural network Deep-CNN structure
Step four: training data generation model
The method comprises the steps of extracting and evaluating the characteristics of the writing quality of the Chinese characters, such as the structural characteristics, the morphological characteristics, the gravity center characteristics, the stroke characteristics, the radical characteristics, the similarity characteristics of standard template characters and the overall writing quality characteristics of the Chinese characters, through training a large number of handwritten Chinese character sets marked with information.
Step five: accuracy of test model
A portion of the word stock is collected and labeled in the first step as a test set to check the accuracy of the model obtained by training.
Step six: method for evaluating Chinese character writing quality by using model
The Chinese characters written by the user are extracted, and then the scores and evaluations of the writing quality of the Chinese characters, such as the scores and the writing evaluations of the structure, the shape, the gravity center, the strokes, the components, the similarity with the standard template characters and the like, and the overall writing scores and evaluations of the Chinese characters can be obtained through the model.
Through Deep learning of the Deep convolutional neural network Deep-CNN structure, the AI can be more fully contacted with more characters and graphs, and the characters and the graphs are obtained through a large amount of collection, so that the comparison of a user in learning is more detailed, and the character evaluation is realized through big data.
Detailed Description
The present invention is described in detail below: a Chinese character writing quality evaluation method based on deep learning comprises the following steps:
the method comprises the following steps: collecting hand-written Chinese characters and establishing character library
The method comprises the steps of collecting general Chinese character handwritten word graphs with different writing qualities as much as possible, and then labeling each word graph, such as writing quality scores, character structure scores, form scores, gravity center scores, stroke scores, radical scores, similarity scores with template standard characters, structural comments, form comments, re-written comments, stroke comments, radical comments, similarity comments with templates, Chinese character writing overall comments and the like.
Step two: data extension based on generation of antagonistic neural network GAN
On the basis of the above small scale with labeled data sets, a semi-supervised deep learning model is used. Through a method of combining the recurrent neural network and the generative confrontation network GAN, the recurrent neural network learns the labeling relation and the characteristics of the data, and the generative confrontation network generates GAN reasonable data so as to expand a data set. And then, through data processing and other work, a reliable data set for model training is formed, and the problem of shortage of the data set is relieved and supplemented. Thereafter, a Deep learning convolutional neural network Deep-CNN structure is used;
step three: deep learning convolutional neural network Deep-CNN structure
Step four: training data generation model
The method comprises the steps of extracting and evaluating the characteristics of the writing quality of the Chinese characters, such as the structural characteristics, the morphological characteristics, the gravity center characteristics, the stroke characteristics, the radical characteristics, the similarity characteristics of standard template characters and the overall writing quality characteristics of the Chinese characters, through training a large number of handwritten Chinese character sets marked with information.
Step five: accuracy of test model
A portion of the word stock is collected and labeled in the first step as a test set to check the accuracy of the model obtained by training.
Step six: method for evaluating Chinese character writing quality by using model
According to the invention, a large amount of written characters are collected, the computer AI can carry out self deep learning, the writing habits of most people can be obtained by calculating big data after learning, most people feel that the characters are beautiful, a relatively complete conclusion can be obtained by accumulating all the year round, and more people can know the self character grading condition by comparing the conclusion with the existing characters.