深度学习方法在地图点注记配置中的应用研究  

Research on the Application of Deep Learning Method to Map Point Annotation Configuration

作　　者：李想[1] 李宏伟[1] LI Xiang;LI Hongwei(School of Geoscience and Technology,Zhengzhou University)

机构地区：[1]郑州大学地球科学与技术学院

出　　处：《出版与印刷》2022年第2期49-56,共8页Publishing & Printing

基　　金：国家自然科学基金项目(编号45171394);教育部产学合作协同育人项目(编号201901024025);郑州大学高层次人才科研启动项目(编号135-32310276)。

摘　　要：为提升机器地图制图的智能化程度,提高地图点注记配置的效率及质量,文章提出基于深度学习的地图点注记配置方法。首先从公开出版的地图集中获取图片构建地图点注记数据集,然后采用卷积神经网络和密集卷积网络模型,基于Tensor Flow和MXNet两种框架,对地图点注记进行类别识别和文字识别,最后根据文字识别结果进行注记位置匹配,并结合注记类别进行注记配置,以实现深度学习方法在地图点注记配置中的应用。实验结果显示,这种方法能够实现地图点注记自动化配置,有效提高地图制图注记配置效率。To improve the intelligence of machine mapping and the efficiency and quality of map point annotation configuration, a map point annotation configuration method based on deep learning is proposed in this paper.Firstly, the paper obtains the pictures from the published altases to construct the map point annotation data sets, and then uses the Convolutional Neural Network and the Dense Convolutional Neural Network model to carry out research from the two aspects of map annotation category recognition and character recognition based on Tensor Flow and MXNet. Finally, the annotation position matching is performed according to the text recognition results, and the annotation configuration is performed in combination with the annotation categories, which realizes the combination of deep learning method and map point annotation configuration. The research method of this paper can realize the automatic configuration of map point annotation and effectively improve the efficiency of map annotation configuration.

关 键 词：地图注记分类 深度学习 文字识别 注记配置 卷积神经网络 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程]