摘要
为了提高无人机影像建筑物自动提取精度,本文采用语义分割U-Net模型完成对无人机正射影像中建筑物的提取,针对目标区域数据太少等问题,使用深度迁移学习的方法在开源数据集训练U-Net模型,然后通过迁移学习对无人机正射影像中的建筑物进行提取。实验采用法国国家信息与自动化研究所(INRIA)公开的图像数据集,以验证实验方法的有效和稳定。实验结果显示,该方法从影像中提取建筑物的准确率为97.36%、精确率为91.00%、召回率为88.51%,并且不受建筑物分布密度和类型的限制,对矩形建筑物提取效果较好。
In order to improve the automatic extraction accuracy of buildings in UAV image,U-Net network is used to complete the extraction of buildings in UAV orthophoto image.To address the problem of too little data in the target area,the deep transfer learning method is used to train the U-Net network in the open source dataset,and then the buildings in the UAV orthophoto image are extracted through migration learning.The experiments are conducted by using the image dataset published by INRIA to verify the validity and stability of the experimental method.The experimental results show that the method can extract buildings from the images with 97.36%accuracy,91.00%precision and 88.51%recall.The method is not limited by the distribution density and type of buildings,and is effective in extracting rectangular buildings.
作者
邓烨
丁涛
DENG Ye;DING Tao(School of Spatial Information and Surveying and Mapping Engineering,Anhui University of Science and Technology,Huainan 232001 China;Key Laboratory of Aviation-Aerospace-Ground Cooperative Monitoring and Early Warning of Coal Mining-Induced Disasters of Anhui Higher Education Institutes,Huainan 232001 China)
出处
《科技创新与生产力》
2021年第7期71-74,共4页
Sci-tech Innovation and Productivity
作者简介
邓烨(1995-),女,江苏无锡人,在读硕士,主要从事摄影测量与遥感研究,E-mail:929883823@qq.com。
