| 多尺度特征融合深度学习建筑物的提取方法 |
| |
| 引用本文： | 刘恒恒,张春森,葛英伟,史书. 多尺度特征融合深度学习建筑物的提取方法[J]. 地理空间信息, 2022, 20(2): 97-100. DOI: 10.3969/j.issn.1672-4623.2022.02.020 |
| |
| 作者姓名： | 刘恒恒 张春森 葛英伟 史书 |
| |
| 作者单位： | 西安科技大学测绘科学与技术学院,陕西 西安 710054 |
| |
| 基金项目： | 陕西省自然科学基金资助项目;国家自然科学基金 |
| |
| 摘 要： | 提出一种基于多尺度特征融合的建筑物提取方法,结合新的网络DenseASPP-UNet,以实现影像多尺度特征的融合,进而高精度提取建筑物.通过Inria开源建筑物航空影像数据集进行验证,表明DenseASPP-UNet相比其他深度学习方法建筑物提取精度有很大的提升.
|
| 关 键 词： | 深度学习 建筑物提取 多尺度特征融合 密集空洞空间金字塔池化 |
Multi-scale Feature Fusion Deep Learning-based Building Extraction Method |
| |
| LIU Hengheng,ZHANG Chunsen,GE Yingwei,SHI Shu. Multi-scale Feature Fusion Deep Learning-based Building Extraction Method[J]. Geospatial Information, 2022, 20(2): 97-100. DOI: 10.3969/j.issn.1672-4623.2022.02.020 |
| |
| Authors: | LIU Hengheng ZHANG Chunsen GE Yingwei SHI Shu |
| |
| Affiliation: | (College of Geomatics,Xi'an University of Science and Technology,Xi'an 710054,China) |
| |
| Abstract: | We proposed a building extraction method based on multi-scale feature fusion.Based on the UNet model,we jumped layers in parallel with the dense cavity space pyramid pooling layer and used 1×1 convolution to reduce feature mapdimension.We proposed a new network Den-seASPPUNet to achieve multi-scale imagefeature fusion and extracted buildings with high precision.We verified it with the Inriaopen source building aerial image data set.The experiment result shows that DenseASPP-UNetgreatly improves building extraction accuracy compared with other deep learning methods. |
| |
| Keywords: | deep learning building extraction multi-scale feature fusion dense cavity space pyramid pooling |
| 本文献已被 维普 万方数据 等数据库收录！ |
