Student handwritten mathematical formula recognition model combining EfficientNet and GNN
Jimin Guo · Systems and Soft Computing · 2025
Recognizing handwritten mathematical formulas in student test papers can greatly increase the scoring speed. However, due to the instability of handwritten mathematical formulas, such as irregular writing and character distortion, the recognition effect of formulas is greatly reduced. Therefore, the research proposes a student handwritten mathematical formula recognition model combining efficient networks and graph neural networks, which uses efficient networks for mathematical formula region detection. Subsequently, the graph neural network performs mathematical formula character recognition, ultimately completing formula recognition and display. The experimental results showed that the recognition rate, recall, accuracy, and harmonic mean of the proposed model for region detection were 96.87%, 96.34%, 95.98%, and 96.45%, respectively. The false alarm rate during character recognition was only 2.12%. The overall recognition rate of handwritten formulas was 56.90% for expression recognition and 69.72% for structure recognition, both of which outperformed the comparison methods. Overall, the handwritten mathematical formula recognition model can enhance the recognition rate. It can effectively address the diversity and complexity in handwritten mathematical formulas, thereby improving the efficiency of paper revision and further promoting the development of educational intelligence.