Test Paper Score Recognition System Based On Deep Learning

Paper test paper has the advantages of easy to use,low cost and so on.As an examination carrier,it is widely used in all kinds of examinations.However,at present,most paper test papers need manual marking and co-marking in practical application.The traditional method is to record the score of each question directly in the score column of the test,and then rely on manual method to synthesize the score of all the sub-items.When the amount of queries is considerable,this technique is prone to errors..In addition,when analyzing the degree of achievement of teaching objectives in schools,it is necessary to manually input the score value of each question,which not only consumes a lot of time and energy of teachers,but also has low efficiency.Therefore,a paper score recognition system based on deep learning is designed and implemented in this paper.The system can automatically recognize the column scores and subjective scores in the paper,and can combine the total score of the paper.Using this system can greatly reduce the workload of manual statistics and reduce the error of manual marking effectively when the examination paper is graded and the achievement degree of teaching objectives is analyzed.Based on deep learning and digital image processing technology,this paper uses real-time sampled test paper pictures as the research carrier for score column positioning,subjective question score positioning and handwritten score recognition.The main research contents include the following three aspects:Firstly,from the traditional method and deep learning method,the paper scores column positioning and subjective question score positioning are studied.Among the traditional algorithms,Huffman frame and line detection method is the most commonly used,but it has some problems such as large amount of computation,high spatial complexity and low efficiency when there are many pixels in the image.From the perspective of design ideas,object detection algorithms based on deep learning can be divided into two categories: one is the single-stage object detection algorithm represented by YOLO series algorithm,the other is the two-stage object detection algorithm represented by R-CNN.In this paper,a method based on YOLOv5 is proposed to quickly locate score column and subjective question score values.This method locates score column and subjective question score values through self-made data set.Secondly,the score bar image and subjective question score value image after positioning and clipping are preprocessed to improve the image quality.Through contour extraction and screening and image normalization processing,the handwritten scores on the test paper can be effectively extracted,and the numerical recognition model based on Le Net-5 can be used for recognition on the MNIST dataset.The network model can not only meet the requirements of universality,but also has a good recognition effect.Finally,the paper provides the overall architecture of the system,including the system block diagram,software and hardware environment and the interface design realized by Py Qt5.This interface can realize the recognition function of test paper scores,and the recognition of the scores after the sum calculation.In this paper,the control variable method is adopted to carry out group test experiments,and the overall accuracy rate of the system is taken as the main evaluation index to carry out experiments on the overall system combining the above two modules.The results show that the system designed in this paper has a good recognition effect for test paper scores.