基于深度学习的电离层数据识别与表格恢复
Ionospheric data recognition and form recovery based on deep learning
鲁月 1苏桂昌 1邵巍 1刘祥鹏1
作者信息
摘要
针对电离层印刷体历史数据清晰度低和字符相似问题,提出基于DBNET与CRNN的电离层数据检测与识别方法.通过图片预处理去除噪声干扰,采用DBNET定位目标图片的文本区域;增加CRNN中的卷积层,将两个池化层的窗口大小改为2 ×1,从而提取更大宽度的特征识别检测到的文本,并加入批量归一化处理加速训练;利用RARE完成识别结果的表格恢复.实验结果表明,该方法对电离层印刷体数据具有良好的识别效果,字符正确识别率达到98％以上.
Abstract
To address the problems of low clarity and character similarity of historical data of ionospheric prints,an ionospheric data detection and recognition method based on DBNET and CRNN is proposed.The noise interference is removed through image preprocessing,and DBNET is used to locate the text area of the target image.The convolutional layer in CRNN is added,and the window size of the two pooling layers is changed to 2 × 1 to extract the text detected by feature recognition with a larger width,and batch normaliza-tion processing is added to accelerate training.RARE is used to complete the form recovery of identification results.Experiment results show that this method has a good recognition effect on ionospheric printed body data,and the correct recognition rate of characters reaches more than 98％.
关键词
深度学习/印刷体数据/字符检测/字符识别/表格恢复Key words
deep learning/printed data/character detection/character recognition/Form recovery引用本文复制引用
基金项目
青岛科技大学2022年国家级大学生创新训练计划项目(202210426009)
出版年
2024
NETL
NSTL
万方数据