教育信息化背景下改进Bi-LSTM的英语手写识别理论研究

Research on Improving English Handwriting Recognition Using Bi-LSTM in the Context of Educational Informatization

摘要: 教育信息化背景下，大规模纸笔考试英语填空主观题依赖人工阅卷，存在效率低、评阅主观偏差大等问题。离线手写文本识别(HTR)是搭建智能阅卷系统的核心技术，CNN-BiLSTM-CTC为英文手写识别主流基础架构，但单层Bi-LSTM难以捕捉长单词远距离字符依赖，且无法自适应区分涂改、连笔等卷面噪声。本文以中学英语统考阅卷为应用场景，提出理论优化方案，设计双层堆叠Bi-LSTM搭配后置注意力机制的改进网络架构，可为基础教育智慧英语测评体系建设提供理论参考。

Abstract: In the context of educational informatization, large-scale paper-based English cloze tests and subjective questions rely on manual grading, which suffers from low efficiency and significant subjective bias. Offline Handwritten Text Recognition (HTR) is the core technology for building intelligent grading systems. CNN-BiLSTM-CTC is the mainstream architecture for English handwritten recognition, but a single-layer Bi-LSTM struggles to capture long-distance character dependencies in long words and cannot adaptively distinguish between corrections, connected strokes, and other form noise. Taking middle school English standardized exam grading as the application scenario, this paper proposes a theoretical optimization scheme and designs an improved network architecture with a double-layer stacked Bi-LSTM combined with a post-attention mechanism, which can provide a theoretical reference for the construction of an intelligent English assessment system in basic education.