@inproceedings{he-etal-2023-rong,
    title = "融合多粒度特征的缅甸语文本图像识别方法({B}urmese Language Recognition Method Fused with Multi-Granularity Features)",
    author = "He, Enyu  and
      Chen, Rui  and
      Mao, Cunli  and
      Huang, Yuxin  and
      Gao, Shengxaing  and
      Yu, Zhengtao",
    editor = "Sun, Maosong  and
      Qin, Bing  and
      Qiu, Xipeng  and
      Jiang, Jing  and
      Han, Xianpei",
    booktitle = "Proceedings of the 22nd Chinese National Conference on Computational Linguistics",
    month = aug,
    year = "2023",
    address = "Harbin, China",
    publisher = "Chinese Information Processing Society of China",
    url = "https://preview.aclanthology.org/Author-page-Marten-During-lu/2023.ccl-1.11/",
    pages = "124--134",
    language = "zho",
    abstract = "{\textquotedblleft}缅甸语属于东南亚低资源语言,缅甸语文本图像识别对开展缅甸语机器翻译等任务具有重要意义。由于缅甸语属于典型的字符组合型语言,一个感受野内存在多个字符嵌套,现有缅甸语识别方法主要是从字符粒度进行识别,在解码时会出现某些字符未能正确识别而导致局部乱码。考虑到缅甸语存在特殊的字符组合规则,本文提出了一种融合多粒度特征的缅甸语文本图像识别方法,将较细粒度的字符粒度和较粗粒度的字符簇粒度进行序列建模,然后将两种粒度特征序列进行融合后利用解码器进行解码。实验结果表明,该方法能够有效缓解识别结果乱码的现象,并且在人工构建的数据集上相比{\textquotedblleft}VGG16+BiLSTM+Transformer{\textquotedblright}的基线模型识别准确率提高2.4{\%},达到97.35{\%}。 ''"
}