Research And Application Of Card-free Examination Paper Layout Analysis And Answer Result Extraction Algorithm

In the field of education,how to facilitate teachers’examination paper grading work and simplify the examination process of reviewers have always been a matter of great concern to the industry.At present,a relatively successful case is the automatic grading system based on optical mark recognition,which can realize automatic grading of objective questions and facilitate teachers’grading of examination papers.However,this way has the disadvantage of high cost of making answer cards and not suitable for small-scale examinations.In order to solve the above problems,more and more researchers have carried out research on the card-free grading system.The card-free grading system can get rid of the dependence of auto grading on answer cards,eliminate the cost of making,scanning answer cards and related supporting equipment,so it can be applied to various scenarios such as small exams and homework,and is the development direction of intelligent education in the future.In this study,three tasks have been done:the analysis of examination paper layout,the extraction of students’answer results and the generation of electronic marked test papers in the card-free grading system.First the examination paper layout analysis dataset and the answer result extraction dataset are constructed based on the papers collected from multiple schools.Then,in the examination paper layout analysis task,this paper proposes a paper layout analysis model based on Swin Transformer optimization,which introduces more contextual features;optimizes the ratio of anchor boxes to make the model more suitable for examination paper document elements;by introducing Mask Io U detection head and OHEM strategy effectively improves the layout analysis effect of the examination paper.In the task of extracting students’answer result,in view of the fact that the characteristics of answering results are mainly shallow features,an effective answering result extraction model is proposed.It is based on the multi-level feature fusion design and optimizing the Unet network structure.Then,based on the content matching algorithm,the above mentioned paper layout analysis results and the students’answer extraction results are matched to generate an electronic marked test paper for card-free grading system.At Last,based on the B/S architecture,An electronic examination paper management system is implemented.Experiments show that compared with the existing methods,the detection performance evaluation index of the examination paper layout analysis model based on Swin Transformer m AP₇₅^bbox increased by 8.7%,and the segmentation performance evaluation index m AP₇₅^segm increased by 12.7%.The optimized answer result extraction model increased the pixel classification accuracy of the answer result category by 13.8%and the intersection ratio by 12.1%.The electronic marked test paper generated based on the content matching algorithm can also meet the requirements of the subsequent identification and correction tasks of the card-free grading system.