|
| 基于深度学习的遥感影像城市苫盖提取 |
| Extraction of Remote Sensing Images of Cites Covered with Dust Suppression Net Based on Deep Learning |
| |
| DOI： |
| 中文关键词: DeepLabv3+模型 神经网络 遥感影像 城市苫盖 |
| 英文关键词: DeepLabv3+ model Neural network Remote sensing image Dust suppression net |
| 基金项目:国家重点研发计划基金资助项目（2021YFB3901103） |
|
| 摘要点击次数: 153 |
| 全文下载次数: 88 |
| 中文摘要: |
| 基于高分一号（GF-1）和高分六号（GF-6）卫星高分辨率遥感影像，利用深度学习DeepLabv3+模型实现城市苫盖识别，并与U型网络（U-Net）、分割网络（Seg-Net）、金字塔场景解析网络（PSP-Net）等方法进行对比。结果表明：城市苫盖样本的最佳裁剪尺寸为256像素×256像素，裁剪方式为随机裁剪；DeepLabv3+模型对苫盖识别的准确率为9840%，召回率为9808%，交并比（IoU）为9654%，均优于U-Net、Seg-Net、PSP-Net等方法；采用同一台服务器进行测试，DeepLabv3+模型运行时间与其余3种方法在同一水平。 |
| 英文摘要: |
| This study utilized the deep learning model of DeepLabv3+ to identify high resolution remote sensing images of cities covered with dust suppression net by GF 1 and GF 6 satellites. The method was compared with the methods of U shaped network（U Net）， segmentation network（Seg Net） and pyramid scene parsing network（PSP Net）， the results showed that the optimal cropping size for samples of images with dust suppression net was 256 pixels × 256 pixels，and the cropping method was random dropping. The DeepLabv3+ model had an accuracy rate of 9840%， a recall rate of 9808% and an IoU ratio of 9654% for dust suppression net recognition， all of which were superior to those by U Net，Seg Net and PSP Net. The test results on the same server indicated that the running time of DeepLabv3+ model was at the same level as the other three methods. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
