Research On Intelligent Recognition And Correction Algorithm Of Mathematical Oral Test Questions Based On Deep Learning

Elementary arithmetic exercises lay the foundation for students’ mathematical development,fostering their thinking habits,enhancing problem-solving abilities,elevating their mathematical literacy,and bolstering competitiveness.Improvement in mental calculation hinges on copious practice,which inevitably generates a wealth of arithmetic problems and test papers.Manually grading these problems demands considerable time and effort,precluding timely feedback to students and hampering their learning outcomes.Moreover,grading accuracy can be compromised by graders’ fatigue,leading to errors.To address these issues,this paper presents an intelligent algorithm based on deep learning computer vision techniques for the automatic detection,recognition,and grading of arithmetic problems,thereby streamlining grading efficiency,alleviating teachers’ and parents’ workload,and averting manual grading errors.Given that the arithmetic problems’ input images may vary in type,illumination,background complexity,and unrelated interference,and that printed and handwritten fonts coexist in arithmetic problems with enormous variations among students’ handwriting,this paper’s primary research endeavors are as follows:1.Construct an arithmetic problem dataset by analyzing real-world input scenarios,collecting image data through smartphone photography,scanning,and web crawlers,and devising a training sample generation method to ensure model training scale while mitigating manual annotation burdens.2.Investigate a text-line detection and localization algorithm for arithmetic problems by comparing and analyzing various deep learning text detection methods,and devise a text detection model based on the DBNet network.To circumvent potential interference,a deeper image feature extraction network,ResNet-50,is incorporated to extract multi-dimensional features from the problem images.Simultaneously,by enhancing semantic information and spatial positioning information to improve the feature representation ability of the network,it can more effectively extract the features of arithmetic problems and avoid interference from irrelevant noise.3.Examine a text-line recognition algorithm for arithmetic problems.Given the diversity of character styles and sizes in problem images,a DenseNet network with four Dense Blocks is employed for feature extraction.BiLSTM is used to integrate features from the convolutional layers,capturing the sequential relationships among characters in the text and enhancing model performance.Lastly,a CTC decoding model is introduced,resulting in an end-to-end problem image text recognition model.Ultimately,after completing the text detection and recognition models,an intelligent arithmetic problem grading algorithm is designed and implemented.Experiments with real-world arithmetic problem images demonstrate that the researched intelligent recognition and grading algorithm is applicable to practical engineering tasks.