Algorithm Design Of Intelligent Marking System Based On Deep Learning

At present,most teachers still use manual scoring method,which consumes a lot of manpower and material resources.With the rapid development of artificial intelligence and computer vision technologies,more and more researchers are focusing on the realization of intelligent marking in order to solve various problems existing in manual marking,while actively responding to the government’s vision of being a powerful country in science and technology.Based on this background,this thesis studies the algorithm design of intelligent marking systems based on deep learning and is committed to providing feasible solutions for the field of intelligent marking.The main contribution of this thesis is summarized as below.Firstly,this thesis establishes two dedicated datasets for the intelligent marking system.One collects local primary school students’ homework and test papers as samples to make datasets for locating arithmetic problems in pictures and identifying characters in arithmetic problems.The other uses an unsupervised method to create a dataset dedicated to the removal of print,which lays a good foundation for better recognition of the handwritten area required.Secondly,taking primary school students’ oral arithmetic problems as an example,this paper proposes an intelligent marking algorithm based on YOLOv3.It takes photos of arithmetic problems and calls the YOLOv3 algorithm twice,locates the oral arithmetic questions in the picture and then extracts.Next,characters are recognized and numerical operations are performed.It can find out the arithmetic questions with incorrect calculations in the original picture and mark the correct answers.The experimental results show that the algorithm can achieve higher recognition accuracy and faster recognition speed.Finally,when teachers score students’ test papers,there are many handwritten answers of students,but there is only one correct answer.Based on this background,this thesis proposes an intelligent marking algorithm based on generative adversarial network and attention model.This design does not need to identify problems and just needs to create the correct answer templates.It innovatively uses the idea of removing raindrops to remove unnecessary printed characters and introduce attention mechanism.It makes the network after adversarial learning to generate pictures which remove prints and leave the handwriting only.Then the handwriting is recognized by the object detection algorithm and the recognition result is compared with the answer template to determine whether it is right or wrong.In this thesis,an intelligent marking algorithm based on YOLOv3 and that based on generative adversarial network and attention model are proposed.After experiments,these two algorithms have achieved good results,providing constructive suggestions for the field of intelligent marking.