Research On Recognition And Correction Algorithm Of Subjective Questions In Primary School Mathematics Based On Deep Learning

In the traditional examination scene,the teacher designs the questions,the examinee answers,and finally the teacher corrects them.The most tedious part of the whole process is the manual marking.Manual marking not only brings heavy workload to teachers,but also easily leads to the problems such as correcting errors and long marking cycle,which has a negative impact on teaching.With the development of computer technology,the automatic identification technology of objective question filling has been widely used.However,subjective questions still rely on manual correction because they involve complex problems such as image handwritten character detection,recognition and understanding.The handwritten text of the examinee and the printed text of the examination paper are distributed in the same picture,so it is necessary to make an accurate distinction in the detection,and at the same time,it is necessary to detect the topic type of the text,so that different grading strategies can be used according to different topic types in the follow-up.In addition,the handwriting of different examinees is different,and the text is compact or sparse,which makes the text detection and recognition face great challenges.In order to solve the problems and challenges such as the difficulty to distinguish the adjacent text lines,the error prone recognition of the same adjacent characters and the variety of topic types,this thesis proposes a multi-question handwritten text detection algorithm based on semantic segmentation and an image handwritten text recognition algorithm based on recurrent neural network.This thesis designs and implements an automatic correction algorithm for different types of questions.Firstly,the data set for text detection and text recognition is constructed based on the existing real primary school mathematics test answer card images and related public data sets;secondly,the related algorithms for detection,recognition and correction are implemented,and the training and testing are carried out on the data set constructed.In this thesis,we have done some experiments on the data set constructed to verify our handwritten text detection,text recognition and automatic correction algorithms.The experimental results show that our methods have higher accuracy,and have practical value,and meets the expected research objective.