基于自适应标注样本的高分辨率影像道路提取  

Road Extraction of High-spatial Resolution Remote Sensing Imagery Based on Automatic Sample Labeling Method

作　　者：袁欢欢 隋立春[1] 徐家利 李彦东 李冠宇 YUAN Huan-huan;SUI Li-chun;XU Jia-li;LI Yan-dong;LI Guan-yu(College of Geological Engineering and Geomatics, Chang'an University, Xi'an 710061, China)

机构地区：[1]长安大学地质工程与测绘学院,西安710061

出　　处：《科学技术与工程》2022年第5期1981-1987,共7页Science Technology and Engineering

基　　金：国家自然科学基金(41372330)。

摘　　要：针对现有道路提取算法中难以大规模人工标注样本类别标签的问题,提出了一种基于自适应标注样本提取遥感影像道路的方法。首先,通过改进的模糊C均值聚类算法提取道路区域,进行初步的样本标注;其次,利用基于二次投票的集成去噪算法定位标签噪声样本,更新样本数据集;再次,将更新后的样本集投入随机森林训练并预测影像的分类结果;最后,对道路提取结果进行多方向形态学滤波去除非道路区域,得到精确的道路提取结果。通过不同分辨率、不同场景、不同方法的实验结果表明,所提方法可以自主选择并标注样本,相比传统算法具有较高的提取精度,对于高分辨率遥感影像中直线型、曲线型道路均有较好的道路提取效果。The existing road extraction algorithms cannot label large-scale samples manually.A method based on automatic sample labeling was proposed to extract road from remote sensing imagery.Firstly,the improved fuzzy C-means clustering algorithm was used to extract the road area and label the samples.Secondly,the label noise samples were located and the sample data set was updated by using the integrated denoising algorithm based on second voting.Thirdly,the updated sample set was put into random forest training and the classification result was predicted.Finally,the road extraction result was filtered by multi-directional morphological filtering to remove the non-road areas,and the accurate road extraction result was obtained.The experimental results of different resolutions,different scenes and different methods show that the proposed method can select and mark samples by itself.Compared with the traditional algorithm,it has higher extraction accuracy,and has better road extraction effect for linear and curved roads in high-resolution remote sensing images.

关 键 词：道路提取 样本标注 投票去噪 随机森林 形态学 高分辨率遥感影像 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]