没有合适的资源?快使用搜索试试~ 我知道了~
首页N-FINDR算法加速策略:体积计算与纯度导向改进
N-FINDR算法加速策略:体积计算与纯度导向改进
需积分: 9 4 下载量 46 浏览量
更新于2024-09-09
收藏 189KB PDF 举报
N-FINDR算法是计算机视觉领域中一种非常流行的分割算法,因其自动化特性以及高效性能而受到广泛关注。然而,其核心挑战在于大量的体积计算、初始随机选择元素成员(EMs)以及无目标的搜索过程,这些因素导致了N-FINDR算法的执行速度较慢,限制了其在实际应用中的广泛采用。 本文主要关注的是如何提升N-FINDR算法的速度,提出两种关键策略来优化算法效率。首先,传统方法中体积计算占据了大量计算资源,作者建议通过替换体积计算为距离计算来降低计算负担。这一步旨在减少不必要的像素比较,显著减少算法的运算成本,从而加快处理速度。 其次,另一个重要的优化措施是根据像素纯度可能性(Pixel Purity Index, PPI)的概念对数据集进行重新组织。PPI是一个衡量像素与特定区域一致性程度的指标,通过利用这个概念,可以更智能地选取初始EMs,确保它们代表的是具有较高纯度的区域。这样不仅提高了初始分割的质量,还减少了盲目的搜索环节,使得后续的搜索过程更加有效率。 作者通过数值实验验证了这些改进措施的有效性。通过结合距离计算优化和基于PPI的选择策略,N-FINDR算法的运行速度得到了显著提升,这对于那些对实时性和处理大规模图像有高要求的应用场景来说,无疑是一项重要的优化成果。这篇文章提供了一种实用的方法,帮助用户在保持N-FINDR算法核心优势的同时,显著提高其在实际应用中的执行效率。
资源详情
资源推荐
Journa2 of Harbin Institute of Technolo (New Series),Vo!.15,No.1,2008
Speed-up for N-FINDR algorithm
WANG Li guo ,ZHANG Ye
王 立 国 . 张 晔
(1.Dept.of Information Engineering,Harbin Institute of Technology,Harbin 150001,China;
2.College of Information and Conmunications Engineering,Harbin Engineering University,Hanbin 150001,China)
Abstract:N—FINDR is a very popular algorithm of endmember(EM)extraction for its automated property and
high出 ciency. Unfortunately,innumerable volume calculation,initial random selection of EMs and blind
searching f0r EMs lead to low speed of the algorithm and limit the applications of the algorithm.So in this paper
two measures are proposed to speed up the algorithm.One of the measures is substituting distance calculation
for volume calculation.Thus the avoidance of volume calculation greatly decreases the computational cost.The
other measure is resorting dataset in terms of pixel purity likelihood based on pixel purity index(PPI)concept.
Then.initial EMs can be selected wel1.founded and a fast searching for EMs is achieved.Numerical experi—
ments show that the two measures speed up the original algorithm hundreds of times as the number of EMs is
more than ten.
Key words:endmember extraction;N—FINDR
CLC number:TP75 Docum ent code:
algorithm;PPI algorithm ;spectral unmixing
A Article ID:1005-9113(2008)01-0141-04
In resent years hyperspeetral remote sensing has been
applied in many fields and the corresponding processing
techniques of hyperspectral images(HSI)have been de-
veloped greatly. In HSI,mixed pixels are a mi xture of
more than one distinct substance.an d thev exist for one of
two reasons.First,some different ma terials may occupy a
single pixel for the low spatial resolution of HSI;Second,
distinct materials are combined into a homogeneous mi x-
ture.In these cases,the resulting spectrum wi11 be some
composite of individual spe ctra.One of the most important
HSI processing techniques is spectral unmixing which
aims at analyzing of mi xed pixels. Spectral unmi xing is
the decomposition of a mixed pixel into a collection of dis-
tinct endmembers (EMs) th a set of fractional abun-
dances which indicate the proportion of each EM¨j. As
the basic step of spectral unmi xing,EM extraction is aim-
cial for computing fractional abundances accurately. In
this case.a number of EM extraction techniqu es -43 have
been propo sed over the past decade.
Pixel purity index (PPI)technique ,based on
the geometry of convex sets,is one of the most Success-
ful EM extraction algorithms.The algorithm projects
every data to a large number of random N—dimensional
vectors(called“skewers”)。along which positions are
pointed out. The points that correspond to extrema in
the direction of a skewer are tallied and the pixels with
the highest tallies are considered the purest ones.A re-
duction of dimensionality is first applied to the original
dataset by using the minimum noise fraction (MNF)
transformation.The shortcoming of PPI algorithm is that
it does not identify a final list of EMs. Th e selected
pure pixels should be compared with library spectra.
Another famous EM extraction technique is N.FIN.
DR algorithm L which is a fully automated identifica.
tion of EMs from multidimensional data space,and re.
duction of dimensionality is also necessary . Th e algo.
rithm is described in the next section. In this algo.
rithm ,the most time consuming part is a trial volume
calculation. Random selection of initial EM s is vulnera.
ble of local optimization,and blind searching for final
EMs leads to a very low speed of the algorithm .
In order to get an automated and efficient EMs ex.
traction algorithm ,two important measures are proposed
to speed up N-FINDR algorithm :substituting distance
calculation for volume calculation:resorting dataset in
terms of pixel purity likelihood based on PPI concept.
1 Original N—FINDR Algorithm
After MNF transformation of original data set,the
spectra of pixels in transform ed space are denoted as s£
(t=1,2,…,M).LetNbe the number of dimensional-
ity of these spectra,then they may be modeled in term s
of a linear combination of(N +1)EM vectors e (i=
1,2,… ,Ⅳ +1),i.e.
=
∑ t
= l
N+1
s.t.∑ t =1,0≤ t ≤1,t=1,2,…,M,(1)
Received 2oo5—09 —15.
Sponsored by the National Natural Science Foundation of China(Grant No.60402025 and 60302019)
维普资讯 http://www.cqvip.com
下载后可阅读完整内容,剩余3页未读,立即下载
shenzhiyun0719
- 粉丝: 1
- 资源: 2
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
- 李兴华Java基础教程:从入门到精通
- U盘与硬盘启动安装教程:从菜鸟到专家
- C++面试宝典:动态内存管理与继承解析
- C++ STL源码深度解析:专家级剖析与关键技术
- C/C++调用DOS命令实战指南
- 神经网络补偿的多传感器航迹融合技术
- GIS中的大地坐标系与椭球体解析
- 海思Hi3515 H.264编解码处理器用户手册
- Oracle基础练习题与解答
- 谷歌地球3D建筑筛选新流程详解
- CFO与CIO携手:数据管理与企业增值的战略
- Eclipse IDE基础教程:从入门到精通
- Shell脚本专家宝典:全面学习与资源指南
- Tomcat安装指南:附带JDK配置步骤
- NA3003A电子水准仪数据格式解析与转换研究
- 自动化专业英语词汇精华:必备术语集锦
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