一种基于迁移学习的栅格型图表自动分类方法  

An Automatic Classification Method of Grid Chart Based on Transfer Learning

作　　者：韩冰[1] 王光霞[1] 陈令羽[1] 王慧芳 张蓝天 HAN Bing;WANG Guangxia;CHEN Lingyu;WANG Huifang;ZHANG Lantian(Information Engineering University, Zhengzhou 450001, China;61206 Troops, Beijing 100043, China;61646 Troops, Beijing 100043, China)

机构地区：[1]信息工程大学,河南郑州450001 [2]61206部队,北京100043 [3]61646部队,北京100043

出　　处：《测绘科学技术学报》2021年第1期75-82,共8页Journal of Geomatics Science and Technology

基　　金：国家重点研发计划项目(2017YFB0503500)。

摘　　要：图表自动分类是实现图表内容解译及信息提取的前提。基于Inception V3模型,采用迁移学习的方法,对图表自动分类方法进行了研究,提出了对5类栅格图表进行自动分类的模型。通过与传统的图表自动分类方法的对比试验,测试了在不同学习率参数下模型的分类效果,给出了针对结构化特征的指数衰减学习率的数学表达式。结果表明:采用迁移学习策略,重点改进学习率参数,可以克服传统图表分类方法的不足,较为高效准确地实现对栅格型图表的自动分类工作。The automatic classification of charts is the premise of the content interpretation and information extraction.Based on the Inception V3,an automatic classification model for five kinds of raster charts was proposed by means of transfer learning.Through the contrast experiment with the traditional automatic classification method,the classification performance of this model under different learning rate parameters was tested,and the mathematical expression of exponential decay learning rate for structural features was given.The results indicate that using the transfer learning strategy and improving the learning rate parameters can overcome the shortcomings in the traditional chart classification methods,and realize the automatic classification of raster chart more efficiently and accurately.

关 键 词：深度学习 卷积神经网络 Inception V3模型 迁移学习 图像分类 栅格图表 

分 类 号：P289[天文地球—地图制图学与地理信息工程]