Research And Implementation Of Handwriting Recognition Model Based On Deep Neural Decision Forest

With the further development of artificial intelligence,intelligent city,wisdom,education and other fields has received the widespread attention.In particular,many key artificial intelligence technologies have been applied in the field of intelligent education,including computer vision,text recognition,speech recognition and so on.Text information is one of the important carriers of information communication,and text recognition and detection plays a vital role in wisdom education.In the era of computer Internet did not rise,learning can only be through books and teachers' oral experience.Nowadays,the Internet is an important learning channel that people can study on it throw online systems.But,the content in the online systems require many people work for it,including content input and marking.Recently,it is very important to have a research on character recognition to improve the level of the wisdom education.Nowadays,the most popular research is to recognize the mathematical formulas for complex structure.In this thesis,handwritten text data of mathematical problem solving process is acquired by camera for processing and handwritten recognition.This thesis studies the recognition of handwritten text data in the process of solving mathematical problems,starting from two types of text: sequential text data and complex structured mathematical formula,the recognition of offline handwritten text is studied.The work of the thesis is as follows:According to the sequential text data,the sequential text is split by using its orderliness,and each character is recognized one by one by using a character handwritten recognition network which have used the deep decision forest.The recognition accuracy of this network is 90.34% in the special data set through the experiments.The accuracy is 2.56% higher than the result of the ordinary convolutional neural network.For complex structured mathematical formulas,the traditional single character recognition model cannot be used to recognize them accurately because of the existence of non-sequential two-dimensional structure.This thesis proposed an offline handwritten mathematical formula recognition network which have used the deep decision forest to recognize the whole complex structure.The recognition accuracy of this network is 45.79% in the latest public data set through the experiments.Aimed at solving math problems contained in the one-dimensional sequence of text and two-dimensional structured mathematical formula two characteristics of this text,combining the single character recognition model and the structural formula recognition model,designed for solving math problems of handwritten recognition module,and for the actual project requirements,develop a set of complete homework system,and validates the effectiveness of the system.