[1]冯 澳,李鸿翔,刘子曦,等.遥感影像水体提取模型的新型研究[J].计算机技术与发展,2021,31(增刊):56-61.[doi:10. 3969 / j. issn. 1673-629X. 2021. S. 011]
　FENG Ao,LI Hong-xiang,LIU Zi-xi,et al.Research on Water Extraction Model from Remote Sensing Image Based on Improved Deeplabv3+ Network[J].,2021,31(增刊):56-61.[doi:10. 3969 / j. issn. 1673-629X. 2021. S. 011]
点击复制

遥感影像水体提取模型的新型研究()
分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
31
期数:
2021年增刊
页码:
56-61
栏目:
图形与图像
出版日期:
2021-12-31

文章信息/Info

Title:
Research on Water Extraction Model from Remote Sensing Image Based on Improved Deeplabv3+ Network
文章编号:
1673-629X(2021)S0056-06
作者:
冯 澳1 ; 李鸿翔1 ; 刘子曦1 ; 王艺霏2 ; 伍 煊1 ; 肖子浛1 ; 刘 涛1
1. 四川农业大学 信息工程学院,四川 雅安 625014;?
2. 西南财经大学 经济与管理研究院,四川 成都 611130
Author(s):
FENG Ao1 ; LI Hong-xiang1 ; LIU Zi-xi1 ; WANG Yi-fei2 ; WU Xuan1 ; XIAO Zi-han1 ; LIU Tao1
1. School of Information Engineering,Sichuan Agricultural University,Yaan 625014,China;
2. Research Institute of Economics and Management,Southwestern University of Finance and Economics,Chengdu 611130,China
关键词:
遥感影像; 水体提取; NDWI; Deeplabv3+; 阈值分割; ResNet-101
Keywords:
remote sensing image; water extraction; NDWI; Deeplabv3+; threshold segmentation; ResNet-101
分类号:
TP399
DOI:
10. 3969 / j. issn. 1673-629X. 2021. S. 011
摘要:
如何从遥感影像中准确且高效地提取出水体特征是学者研究的重点。 无论是传统方法还是水体指数计算法,大部分都无法相对统一应用于各种遥感影像中的水体提取,也无法解决城市中细小水体难以被提取的问题,具有一定的局限性。 针对以上问题,文中提出一种基于改进 Deeplabv3+网络的新型提取模型。 首先对遥感影像进行大气校正和假彩色处理,降低数据色差的影响,有效地对水体与背景色进行颜色区分。 再根据 NDWI 计算和阈值分割,得到二值 NDWI 特征图,将波段信息与假彩色图像进行特征融合。神经网络模型选择 Deeplabv3+作为框架结构,并采用 ResNet-101 作为优化的骨干网络。通过对 ASPP 模块改进后,提高了整体提取效果。与选取的 U-Net, SegNet, FCN,PSPNet 四种模型进行对比,改进 Deeplabv3+模型在 PA 上分别提高了 3. 893% ,5. 242% ,4. 318% 和 3. 471% ,在 IOU 上分别提高了 6. 534% ,3.732% ,3. 749% 和 6. 331% ,在 Recall 上分别提高了 6. 111% ,6. 698% ,5. 776% 和 4. 901% 。 同时在提取结果上也降低了噪点,增强了提取水体的完整度和边缘细化能力,具有很好的实用性。
Abstract:
How to accurately and efficiently extract water features from remote sensing images is the focus of scholars’ research. Whether it is traditional methods or water index calculation methods,most of them cannot be relatively uniformly applied to water extraction in various remote sensing images,nor can they solve the problem that small water bodies in cities are difficult to extract, and they have certain limitations. In response to the above problems。 we propose a new extraction model based on improved Deeplabv3 + network.Firstly,atmospheric correction and false color processing are performed on remote sensing images to reduce the influence of data color difference and effectively distinguish the water body from the background color. Then according to NDWI calculation and threshold segmentation,a binary NDWI feature map is obtained,and the band information is feature - fused with the false color image. The neural network model chooses Deeplabv3+ as the frame structure,and uses ResNet-101 as the optimized backbone network. After improving the ASPP module,the overall extraction effect is improved. Compared with the selected U-Net,SegNet,FCN and PSPNet,the improvedDeeplabv3+ model has increased 3. 893% ,5. 242% ,4. 318% ,3. 471% on PA,and 6. 534% ,3. 732% ,3. 749% ,6. 331% on IOU,and6. 111% ,6. 698% ,5. 776% ,4. 901% on Recall. At the same time,the extraction result also reduces noise,enhances the integrity of the extracted water body and the edge refinement ability,and has good practicability.

相似文献/References:

[1]余 威,龙慧云.基于深度卷积网络的遥感影像建筑物分割方法[J].计算机技术与发展,2019,29(06):57.[doi:10. 3969 / j. issn. 1673-629X. 2019. 06. 012]
　YU Wei,LONG Hui-yun.A Building Segmentation Method Based on Deep Convolution Networks for Remote Sensing Imagery[J].,2019,29(增刊):57.[doi:10. 3969 / j. issn. 1673-629X. 2019. 06. 012]
[2]单传辉,叶绍华,姚万琪,等.基于深度可分离残差网络的遥感影像路网检测[J].计算机技术与发展,2023,33(04):75.[doi:10. 3969 / j. issn. 1673-629X. 2023. 04. 011]
　SHAN Chuan-hui,YE Shao-hua,YAO Wan-qi,et al.Remote Sensing Image Road Network Detection Based on Deep Separable Residual Network[J].,2023,33(增刊):75.[doi:10. 3969 / j. issn. 1673-629X. 2023. 04. 011]
[3]王 诏*,王 燕,苏国辉,等.一种阶梯型多尺度神经网络条带噪声降噪模型[J].计算机技术与发展,2023,33(07):68.[doi:10. 3969 / j. issn. 1673-629X. 2023. 07. 010]
　WANG Zhao*,WANG Yan,SU Guo-hui,et al.A Staircase Multi-scale Neural Network Model for Stripe Noise Reduction[J].,2023,33(增刊):68.[doi:10. 3969 / j. issn. 1673-629X. 2023. 07. 010]

更新日期/Last Update: 2021-09-10