| 基于DeepLabv3架构的高分辨率遥感图像分类 |
| |
| 引用本文： | 张鑫禄,张崇涛,戴晨光,季虹良,王映雪.基于DeepLabv3架构的高分辨率遥感图像分类[J].海洋测绘,2019,39(2):40-44. |
| |
| 作者姓名： | 张鑫禄 张崇涛 戴晨光 季虹良 王映雪 |
| |
| 作者单位： | 战略支援部队信息工程大学地理空间信息学院,河南郑州,450000;32023部队,辽宁大连,116021 |
| |
| 摘 要： | 针对目前使用机器学习解决高分辨率遥感图像分类主要存在下采样导致的细节信息丢失问题,提出了一种基于DeepLabv3架构的小波域DeepLabv3-MRF(Markov random field,MRF)算法。选择当前较为普遍的DeepLabv3架构分类算法,能够获得更为精确的分类结果;采用小波域DeepLabv3-MRF算法,还能够获得更为清晰的边缘细节信息。选取南方某地区高分辨率无人机遥感图像进行分类实验,通过小波变换的方向性、非冗余性以及MRF变换像素空间的交互性这三个方面,将分类结果与原始DeepLabv3架构分类结果对比分析。结果表明,所提出的分类方法精度明显高于原始DeepLabv3架构分类算法的精度,总体精度可提升3%左右,并且可以充分表达高分辨率遥感图像细节信息。
|
| 关 键 词： | 计算机视觉 深度学习 DeepLabv3架构 高分辨率遥感图像 MRF算法 图像分类 |
High Resolution Remote Sensing Image Classification Based on DeepLabv3 Architecture |
| |
| ZHANG Xinlu,ZHANG Chongtao,DAI Chenguang,JI Hongliang,WANG Yingxue.High Resolution Remote Sensing Image Classification Based on DeepLabv3 Architecture[J].Hydrographic Surveying and Charting,2019,39(2):40-44. |
| |
| Authors: | ZHANG Xinlu ZHANG Chongtao DAI Chenguang JI Hongliang WANG Yingxue |
| |
| Institution: | Institute of Surveying and Mapping,Information Engineering University,Zhengzhou 450000 ,China;32023 Troops,Dalian,Dalian 116021 ,China |
| |
| Abstract: | While applying the method of machine learning to high resolution remote sensing image classification,the loss of detail information caused by downsampling always exists.Therefore,a DeepLabv3-MRF (Markov random field,MRF) algorithm based on DeepLabv3 architecture is proposed to deal with this problem.In this proposed algorithm,we select the DeepLabv3 architecture classification algorithm which is more widely used to obtain more accurate classification results.And the wavelet domain DeepLabv3-MRF algorithm may also help to get more clear edge details.Then we use the high-resolution UAV remote sensing images of an area in south China to carry out classification experiments and compare these results with the original DeepLabv3 architecture classification results in three aspects-the directionality,non-redundancy of the wavelet transform and the interactivity of the MRF transform spatial pixel.The results show that the accuracy of the proposed classification method in this paper is significantly higher than that of the original classification algorithm,the overall accuracy can be improved by about 3%,and the detailed information of high-resolution remote sensing images can be fully expressed. |
| |
| Keywords: | computer vision deep learning DeepLabv3 architecture high resolution remote sensing image MRF algorithm image classification |
| 本文献已被 CNKI 维普 万方数据 等数据库收录！ |
| 点击此处可从《海洋测绘》浏览原始摘要信息 |
| 点击此处可从《海洋测绘》下载免费的PDF全文 |
|
