没有合适的资源?快使用搜索试试~ 我知道了~
首页多字典稀疏表示法在超分辨率映射中的应用
"该资源是一篇研究论文,探讨了如何利用稀疏表示通过多个字典进行超分辨率映射,以提高遥感图像处理的精度和鲁棒性。" 在超分辨率映射(Superresolution Mapping, SRM)领域,研究者们通常依赖于空间依赖性假设来预测混合像素内的土地覆盖类别的空间位置。黄惠娟、于静和孙卫东在2014年发表的这篇论文中提出了一种新颖的方法,即基于多字典的稀疏表示超分辨率映射。这种方法针对学习过程和类别分配过程中可能出现的噪声具有较强的鲁棒性。 在所提出的算法中,首先根据光谱失真的程度来获取每个类别所占的亚像素数量。不同的类别被区别对待,它们的分布模式被视为可区分的。接着,通过学习到的多个分布字典,依据归一化的重构误差进行亚像素分类。这种方法的核心在于利用稀疏表示理论,即用尽可能少的基向量组合来表示原始信号,以提取出数据的内在结构。 实验结果显示,与传统方法相比,该方法在实际图像上的表现具有更高的准确性和鲁棒性。多字典学习的概念是关键,它允许模型更灵活地适应不同的特征模式,从而提高了超分辨率重建的质量。 关键词包括:多字典学习、稀疏表示、空间依赖性和超分辨率映射。这一创新技术对遥感图像分析,特别是土地覆盖分类和遥感图像的解析精度提升有显著作用,有助于环境监测、城市规划等领域的应用。
资源详情
资源推荐
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. 11, NO. 12, DECEMBER 2014 2055
Superresolution Mapping Using Multiple
Dictionaries by Sparse Representation
Huijuan Huang, Jing Yu, and Weidong Sun
Abstract—Superresolution mapping can predict the spatial lo-
cation of land cover classes within mixed pixels based on the spa-
tial dependence assumption. We propose a novel superresolution
mapping method via multidictionary-based sparse representation,
which is robust to noise in both the learning and class-allocation
process. In the proposed method, the subpixel number belonging
to each class is obtained according to the degree of spectral
distortion, and the distribution modes of different classes are
treated discriminatorily. The subpixel classification is performed
according to the normalized reconstruction errors by the learned
multiple distribution dictionaries. The experimental results show
that the proposed method has improved accuracy and robustness
for real imagery.
Index Terms—Multidictionary learning, sparse representation,
spatial dependence, superresolution mapping (SRM).
I. INTRODUCTION
I
N multispectral and hyperspectral images (MHSIs), each
pixel contains a spectral signature over many narrow con-
tinuous spectral bands, which makes the classification of land
cover possible based on their electromagnetic spectrum, not
just the visible spectrum. However, MHSIs are commonly
dominated by mixed pixels that contain more than one distinct
substance due to the intrinsic issues (e.g., imperfect imaging
optics and secondary illumination), as well as extrinsic issues
(e.g., atmospheric turbulence and diverse distributed land cover
classes).
Atkinson [1] first introduced the concept of superresolution
mapping (SRM) based on the assumption of spatial depen-
dence, which refers to the tendency that spatially proximate
observation of a given property to be more alike than more
distant observations. This technique uses the abundance yielded
by spectral unmixing as inputs, and its output is a high-spatial-
resolution (HR) land cover map. By SRM, original low-spatial-
resolution (LR) pixels are divided into several subpixels, and
the ultimate task is to assign each subpixel to a proper class. It
is also termed as subpixel mapping [2], [3]. The current SRM
methods can be roughly categorized into two groups [4]: spatial
Manuscript received November 5, 2013; revised February 25, 2014; accepted
April 8, 2014. This work was supported in part by the National Natural Science
Foundation under Grant 61171117, by the National Science and Technology
Pillar Program under Grant 2012BAH31B01, and by the Key Project of the
Science and Technology Development Program of BEC of China under Grant
kz201310028035.
The authors are with the Department of Electronic Engineering, Tsinghua
University, Beijing 100084, China (e-mail: hhj09@mails.tsinghua.edu.cn;
yujing@tsinghua.edu.cn; wdsun@tsinghua.edu.cn).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/LGRS.2014.2318758
optimization types [2], [3], [5], [6] and learning-based types
[7]–[10]. The first group involves optimizing a spatial regu-
larization model that reflects the spatial dependence, whereas
the second group learns prior information of distribution modes
from known data through neural networks or other learning
methods. In this letter, we focus on the latter category.
The overfitting problem is common in most of the neural-
network-learning-based SRM methods, which are also sensitive
to noise both in the learning and class-allocation process. In
addition, most of the current SRM methods strictly or to a large
extent maintain the number of subpixels belonging to each class
only according to the abundance. More importantly, current
methods usually treat all classes equally and learn only one
distribution mode, which may not be reasonable. For example,
rivers usually distribute linearly, whereas islands distribute as
oval areas, and they should be treated differently.
In natural images, the widely existed many similar patches
can be learned to formulate dictionaries via sparse represen-
tation to do image processing task [11], [12]. Similarly, in
land cover maps, there also exist repetitive spatial structures,
such as linearly distributed rivers. Based on this assumption,
we propose a novel SRM method via multidictionary-based
sparse representation. In this letter, we propose a feature vector
that captures the significant information about spatial depen-
dence and learns the distribution dictionary for each class
via sparse representation. In the class-allocation process, the
feature vector of the underlying subpixel is reconstructed by
every dictionary and is assigned to a class according to the
reconstruction errors and introduced spectrum distortions. Our
framework treats every land cover class discriminatorily, lead-
ing to more accurate distribution dictionaries. It is worth noting
that the sparse representation learning skill is robust to noise,
which has been demonstrated in the face recognition method
[12]. Moreover, the sparse representation has been successfully
applied in classification at pixel scale [13].
The rest of this letter is organized as follows. Section II
presents the proposed method, referred to as MSRSM. The
results of experiments are discussed in Section III. Finally, the
conclusion is drawn in Section IV.
II. P
ROPOSED METHOD
The proposed method can be divided into two steps: learning
and SRM. In the learning step, the training sample sets are first
extracted from the HR training land cover maps, which can
encode the spatial dependence effectively. Then, the multiple
distribution dictionaries are learned via sparse representation.
In the SRM step, the feature vector of the underlying subpixel is
1545-598X © 2014
IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
下载后可阅读完整内容,剩余4页未读,立即下载
weixin_38737176
- 粉丝: 2
- 资源: 928
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
最新资源
- C++多态实现机制详解:虚函数与早期绑定
- Java多线程与异常处理详解
- 校园导游系统:无向图实现最短路径探索
- SQL2005彻底删除指南:避免重装失败
- GTD时间管理法:提升效率与组织生活的关键
- Python进制转换全攻略:从10进制到16进制
- 商丘物流业区位优势探究:发展战略与机遇
- C语言实训:简单计算器程序设计
- Oracle SQL命令大全:用户管理、权限操作与查询
- Struts2配置详解与示例
- C#编程规范与最佳实践
- C语言面试常见问题解析
- 超声波测距技术详解:电路与程序设计
- 反激开关电源设计:UC3844与TL431优化稳压
- Cisco路由器配置全攻略
- SQLServer 2005 CTE递归教程:创建员工层级结构
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功