Research on Automatic Grading of Handwritten Homework in Intelligent Education Based on Model Fusion Approach

With the popularization of intelligent education, more and more handwritten homework is required to be photographed and uploaded on the online platform for submission. Teachers need to grade the handwritten homework submitted by students on the platform. A large number of repetitive grading homework brings a lot of burden to teachers and online grading also causes damage to teachers' eyesight. However, the current research in this area only focuses on the recognition of single font symbols, and cannot effectively and automatically grade the answers submitted by students in the actual teaching environment that contains multiple types of text,symbol and image. To address this problem, we propose an automatic handwritten homework grading system based on a model fusion method to assist teachers in the automatic marking and grading of homework. The model binarizes the collected handwritten tasks to reduce factors such as ambient light. Then the KNN model is used to divide the binarized images into two categories: digital symbols and graphic assignments. The YOLO-CoordAtt model is proposed for graphic homework recognition and automatic grading. The model predicts the target position and category of the scoring points in the answers submitted by students, and scores them according to different step points. Finally, all the step points obtained by the students are added up to output the student's score. In daily teaching environments, there are usually a large number of mathematical symbols in homework. Due to the complex symbol structure and different writing styles, YOLO-CoorAtt is usually difficult to accurately identify them. We proposed CSP-Posformer to solve this problem. CSP-Posformer converts the mathematical symbol homework submitted by students into latex expressions and compares them with the latex expressions of the scoring points entered by the teacher. If they are the same, they will be given scores, realizing automatic grading of handwritten homework containing mathematical symbol formulas. Through these methods, we can quickly and accurately correct the homework submitted by students, thereby freeing teachers from complex and repetitive correction tasks and providing timely feedback to students' submitted answers.