Scholarlink

<p>Police handwritten document recognition plays an increasingly important role in improving new quality combat capability of public security organs. To address the challenges associated with the recognition, this paper proposes a police handwritten document recognition method based on Qwen3VL with EI-LFT using FP-EESR and MFE-GLS. Firstly, Fusion Preprocessing with Edge Enhancement for Super-Resolution Reconstruction(FP-EESR) was used to enhance the handwriting features of handwritten Chinese characters while Multi-scale Feature Extraction using Global-Local-Semantic (MFE-GLS) was employed to extract “global-local-semantic” fusion features to improve recognition performance, as well as Embedding Instruction &amp; LoRA Fine-Tuning (EI-LFT) was utilized to fine tune Qwen3VL to adapt to specific scenarios of police handwritten document recognition. Secondly, to test the recognition and extraction performance of various methods, a dataset was specifically constructed for the recognition and extraction of handwritten police documents. Finally, Multiple comparative experiments were conducted, and theoretical research and experimental results show that the new proposed PHDReader outperforms advanced OCR models and VLM in terms of recognition performance, stability and generalization ability.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

PHDReader: police handwritten document recognition method based on VLM with EI-LFT using FP-EESR and MFE-GLS

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月16日

查看全文
  • Dongxing Wang,
  • Zhiyang Huang

摘要

Police handwritten document recognition plays an increasingly important role in improving new quality combat capability of public security organs. To address the challenges associated with the recognition, this paper proposes a police handwritten document recognition method based on Qwen3VL with EI-LFT using FP-EESR and MFE-GLS. Firstly, Fusion Preprocessing with Edge Enhancement for Super-Resolution Reconstruction(FP-EESR) was used to enhance the handwriting features of handwritten Chinese characters while Multi-scale Feature Extraction using Global-Local-Semantic (MFE-GLS) was employed to extract “global-local-semantic” fusion features to improve recognition performance, as well as Embedding Instruction & LoRA Fine-Tuning (EI-LFT) was utilized to fine tune Qwen3VL to adapt to specific scenarios of police handwritten document recognition. Secondly, to test the recognition and extraction performance of various methods, a dataset was specifically constructed for the recognition and extraction of handwritten police documents. Finally, Multiple comparative experiments were conducted, and theoretical research and experimental results show that the new proposed PHDReader outperforms advanced OCR models and VLM in terms of recognition performance, stability and generalization ability.