随着大语言模型的飞速发展，其在中小学生作文辅导上的潜力日益突出，然而如何精准地将作文图片转化为文本文档仍然是一项极具挑战性的工作，基于深度学习的OCR技术给这一挑战提供了良好的解决方案。本文首先调研了近些年文本检测方向大放异彩的两个模型：DBNet，随后详细分析介绍了在近些年文本识别领域的诸多重大突破：从基于CTC的CRNN，到基于Attention的ASTER，再到基于Transformer的TrOCR。而本文选择了最具代表性的CRNN、TrOCR和hybrid三个文本识别模型进行作文识别的性能评估，为了控制变量三者均采用了DBNet进行文本检测，通过F1指标对比并分析了三个模型在作文识别这一任务上的性能差异。最终验证了OCR技术在中小学生作文字体识别上应用的可行性与有效性。

With the rapid development of large language models, their potential in essay tutoring for middle and primary school students is becoming increasingly prominent. Deep learning-based OCR technology provides a good solution to this challenge. Deep learning-based OCR technology can generally be divided into two parts: text detection and text recognition. This paper first surveys the two models that have shone in text detection in recent years: DBNet and PSENet. It then provides a detailed analysis and introduction of the many major breakthroughs in text recognition in recent years: from the CTC-based CRNN to the attention-based ASTER, to the transformer-based TrOCR. In this paper, the performance of three representative text recognition models, namely CRNN, TrOCR, and hybrid, is evaluated for essay recognition. To control variables, all three models use DBNet for text detection. The F1 index is used to compare and analyze the performance differences of the three models in the task of essay recognition. Finally, the feasibility and effectiveness of OCR technology in recognizing the text of middle and primary school students' essays are verified.