遥感科学与应用技术

基于优化合并的高分辨率遥感影像分割算法

  • 苏腾飞 ,
  • 张圣微 , * ,
  • 李洪玉
展开
  • 内蒙古农业大学水利与土木建筑工程学院,呼和浩特 010018
*通讯作者:张圣微(1979-),男,博士,副教授,硕士生导师,主要从事定量遥感、生态水文及气候变化等方面的研究。E-mail:

作者简介:苏腾飞(1987-),男,硕士,助理实验师,研究方向为面向地理对象影像分析的遥感数据处理算法的设计与实现。 E-mail:

收稿日期: 2015-04-01

  要求修回日期: 2015-12-02

  网络出版日期: 2016-07-15

基金资助

国家自然科学基金项目“科尔沁沙地典型生态系统水热通量传输机理及其与植被耦合关系试验和模拟研究”(51569017)

内蒙古自然科学基金项目“半干旱区沙地典型生态系统水热通量传输机理研究”(2015MS0514)

中国博士后科学基金面上资助“西部地区博士后人才资助计划”(2015M572630XB)

Segmentation Method Using Optimized Merging for High Resolution Remote Sensing Images

  • SU Tengfei ,
  • ZHANG Shengwei , * ,
  • LI Hongyu
Expand
  • Water Conservancy and Civil Engineering College, Inner Mongolia Agricultural University, Hohhot 010018, China
*Corresponding author: Zhang Shengwei, E-mail:

Received date: 2015-04-01

  Request revised date: 2015-12-02

  Online published: 2016-07-15

Copyright

《地球信息科学学报》编辑部 所有

摘要

高分辨率遥感影像的分割算法研究对遥感数据处理与应用具有重要意义。本文提出了一种优化合并的分割算法以提高运算效率,该算法包含局部最优合并和全局最优合并2个阶段。第1阶段采用凝聚层次聚类(Hierarchical Agglomerative Clustering,HAC)方法实现局部最优合并,并对其合并规则进行了优化,使优化后的合并规则先注重光谱特征,再考虑待合并区域的几何特征。第2阶段采用区域邻接图(Region Adjacency Graph,RAG)方法实现全局最优合并,其合并规则主要考虑了区域的光谱和边界信息,减少了区域尺度对合并规则函数产生的负面影响,并且该阶段利用了红黑树来实现全局最优合并,以提高对RAG的搜索效率。最后,利用OrbView3高分辨率遥感影像开展了分割实验,结果表明本文算法可以得到令人满意的分割精度。本文的成果为遥感影像分割及其相关研究提供了新思路。

本文引用格式

苏腾飞 , 张圣微 , 李洪玉 . 基于优化合并的高分辨率遥感影像分割算法[J]. 地球信息科学学报, 2016 , 18(7) : 931 -940 . DOI: 10.3724/SP.J.1047.2016.00931

Abstract

Study on the segmentation method for high resolution remote sensing images is very important for the processing and application of remote sensing data. Image segmentation plays an important role in geographic object-based image analysis, and it is also very useful in GIS data management and remote sensing data compression. A new segmentation algorithm using optimized merging criteria is proposed in this paper. The proposed algorithm divides the merging process into two stages, including the local best merging and the global best merging. Hierarchical agglomerative clustering is used to implement the first stage to meet the main objective of increasing the running efficiency. The merging criterion in the first stage focuses on the regional geometric information to create the visually pleasing segments, and in addition, this criterion is constructed on the premise that the regions to be merged should be sufficiently similar in spectra. Thus, when designing the merging criterion of the local best merge, the spectral and geometric information are both taken into consideration. Moreover, Global Moran′s I is used to determine the ending condition for the first stage. After the local best merging, the region adjacency graph (RAG) is constructed to implement the global best merging, in which the spectral and edge information is taken into account. In this stage, the negative impact introduced by the regions′ scale is found throughout the experiments. Thus, the size information of each region is excluded from the merging criterion of the global best merging. In addition, a special binary search tree, which is called the red-black tree, is used in the implementation to rank the edges of RAG, so as to speed up the graph structure updating after a merging taking place. High resolution images acquired from OrbView3 are adopted to conduct the segmentation experiment, the results of which indicate that our algorithm can produce the satisfactory performance. The conclusions made in this paper may provide new insights for the studies on remote sensing image segmentation and the related researches.

1 引言

随着越来越多高分辨率(High Resolution,HR)遥感卫星(如GeoEye、IKONOS、QuickBird、OrbView)的发射升空,HR遥感影像的信息提取工作面临着严峻的挑战。影像分割作为基于地物影像分析(Geographic Object-Based Image Analysis,GEOBIA)的第一步[1],已被广泛应用于HR遥感数据的自动解译中,包括农业遥感[2]、陆地变化[3]、海岸带制图[4]、军事监测[5]等。而且影像分割在地理信息系统(Geographic Information System,GIS)中扮演重要角色,其可将栅格影像转化为矢量数据,从而方便遥感与GIS的无缝连接,以提高HR数据分析的自动化和实效性。此外,影像分割还有一个重要的应用是影像压缩,可改进影像压缩的效果[6]。随着遥感卫星的不断增多,遥感数据量正以几何级数逐年增长,如何高效地节省存储空间是遥感学界一个不容忽视的问题。因此,开展HR遥感影像分割算法的研究具有非常重要的意义。
目前,主流的HR遥感影像分割算法模型包括:马尔科夫随机场[6]、主动等值线演化模型[7]、分水岭[8]、区域生长[8-12]等。其中,区域生长算法的精度较高,实现较为简单,受到较多学者的关注。早在1989年,Beaulieu和Goldgerg[9]提出了一种基于层次逐步优化(Hierarchical Step-Wise Optimization,HSWO)的区域生长方法来解决影像分割问题。由于HSWO计算量巨大,Kurita[10]和Tilton等[11]对其进行了改进:前者利用一种高级数据结构——堆(heap),来优化HSWO搜索最优合并的速度;后者则实现了HSWO的并行策略,并将其命名为递归层次分割算法(Recursive Hierarchical Segmentation,RHSeg)。然而,以上算法的区域合并规则仅考虑了光谱信息,其分割结果的视觉效果往往不是特别理想。Baatz和Schape[12]提出了一种分形网演化算法(Fractal Net Evolution Algorithm,FNEA),将几何形状特征引入了区域合并规则中,从而显著提升了区域生长算法分割结果的视觉效果。FNEA被集成在商业遥感数据分析软件eCognition中,便于学者进行GEOBIA的相关研究[13-14]
在FNEA的合并规则中,几何特征与光谱特征只进行了简单的线性组合,难以对包含不同地物内容的遥感影像进行分割。近年来,一些学者在FNEA的基础上,提出了若干改进方法。朱俊杰等[15]将边缘信息加入到FNEA分割过程中,以提高分割结果与实际地物的匹配程度;贾春阳等[16]提出了一种自适应权值的FNEA分割算法,以提升合并规则的准确性;Zhong等[17]将FNEA与Multi-agent系统相结合,提出了一种改进的面向对象的遥感影像分割和分类方法;Zhang等[18]也将边缘信息加入到类似FNEA的区域合并规则中,提出利用一种随迭代次数变化的尺度参数来优化合并效果。从合并顺序的角度,Zhang等[19]又提出了一种混合区域合并的方法,来兼顾局部最优合并与全局最优合并的优势,从而提高算法精度。
虽然上述关于FNEA合并规则改进的工作取得了一定的成效,但是在FNEA合并规则函数中,光谱异质性与几何异质性的线性组合关系仍然没有改变。因此,本文在FNEA、HSWO合并规则的基础上进行了优化,试图获取最理想的HR遥感影像分割结果。

2 HR遥感影像分割算法原理

2.1 算法流程

本算法主要分为2步:基于凝聚层次聚类(Hierarchical Agglomerative Clustering,HAC)[20]的局部最优合并,以及基于区域邻接图(Region Adjacency Graph,RAG)的全局最优合并,具体流程如图1所示。本文算法的合并规则有2种,分别是局部最优合并规则 ( C lo cal ) 和全局最优合并规则 ( C global )
Fig.1 Algorithm flowchart

图1 算法流程

2.2 局部最优合并

2.2.1 基于HAC的局部最优合并
HAC是一种空间聚类算法,在海冰与溢油SAR影像的分割中具有较好的效果[20-21]。利用HAC来实现本文算法的第一步是因为:(1)在影像分割的初始阶段,各个像元被当作单独的区域来对待,如果在此时建立RAG,即便是采用最简单的四邻域方式,边的数目也非常大,导致RAG内存占用量大,运算效率低,而基于HAC的分割算法在初始阶段不需记录区域的邻接信息,从而节省了内存空间。(2)HAC注重的是局部最优合并。在初始阶段,每个像元都是单独的区域,这可被认为是极端的过分割。此时,各个相邻区域可被合并的概率较高,局部最优的合并标准可满足精度要求。
HAC将影像划分为形状较规则、纹理较一致的若干区域,这与超像素方法相似。但本文发展的HAC与后续的全局最优合并采用了相同的数据结构,使实现更为简单,而超像素方法采用了完全不同的数据结构,使RAG构建更为复杂,从而降低运算效率。
HAC算法的详细过程可参考文献[20]。需要重点指出的是,本文HAC算法的终止条件是:当影像被足够地过分割后,即开始全局最优合并。本文采用了全局Moran指数(Global Moran′s I,GMI)来进行算法终止的判断。GMI是衡量空间自相关性的统计量,最早由P. Moran提出[22]。对于影像分割,其计算公式为(式(1)):
GMI = n s i = 1 n s j = 1 n s w ij g i - g g j - g i = 1 n s g i - g 2 w ij (1)
式中: n s 是区域数目; g 为图像均值; g i 为区域 i 的均值; w ij 是区域邻接系数,当且仅当区域 i j 邻接时, w ij = 1 ,否则 w ij =0。
GMI的取值范围是0-1,其值越小说明分割结果中各个区域的灰度相差越大。在分割的初始阶段,所有像元都是单独的区域,此时相邻区域之间的灰度差异最小,GMI最大;随着合并的进行,各个区域之间的空间自相关性降低。经过多次实验,发现当HAC局部最优合并使GMI小于0.8时,可以得到令人满意的过分割结果。结合GMI的HAC局部最优合并算法的具体流程如下:
(1)按从上往下、从左往右的顺序访问每一个像元,若该像元被合并,则访问下一个像元,否则进入步骤(2);
(2)对于当前像元所在的区域 S i ,搜索与其邻接的所有区域,并将其存入列表 neb_list ;
(3)在 neb_list 中寻找与 S i 最适合合并的区域,然后将其与 S i 合并(HAC聚类操作);
(4)计算当前分割结果的 GMI ,若其值小于0.8,则终止算法,否则返回步骤(1)。
2.2.2 C lo cal 的设计
本文设计 C local 的准则是:对于2个相邻的区域,只要它们在光谱上足够相似,则说明可以被合并;但为了得到在视觉上较为美观的区域,需考虑形状信息。对于一个区域 S i 及其邻域 S j ,计算它们的光谱异质性变化值,如式(3)所示。
Δ H spec tr al i , j = n m σ m - n i σ i + n j σ j (2)
式中: σ 为区域内部像元光谱值的标准偏差,对于多波段图像,其 σ = ( 1 / B ) Σ b σ b B 为波段数; n 表示区域内部像元数; m 表示区域 i j 合并后的区域。此外式(2)也可用于从多光谱影像中提取的纹理特征影像。若 S i 与其邻域 S j 的光谱异质性变化值 ΔH ( i , j ) 小于阈值 T sl ,则将 S j 视为候选区域。在所有候选区域中,寻找与 S i 的紧凑异质性差异最小的区域,即为将与 S i 合并的区域。紧凑异质性变化值的计算式为式(3)。
Δ H compactness i , j = n m l m - n i l i + n j l j (3)
式中: l 为区域的边界长度。紧凑异质性变化量越小,合并后区域的形状越规则,即越接近圆形,因为圆形具有最小的紧凑性异质性。因此,在合并规则中考虑紧凑性异质性,可得到形状较为规则的斑块。 T sl 的计算式为式(4)。
T sl ( i , j ) = 1 2 B b B σ b l ln ( n i + n j ) (4)
式中: σ b l 为整景图像波段 b 的光谱标准差。式(4)是一个经验公式,可以保证灰度相似的2个区域被合并。式(2)、(4)都考虑了待合并的2个区域的大小信息,这是为了使较小区域之间的合并更容易发生,从而加快算法运算速度。式(4)采用对数函数,是因为这种函数增长率较小。当2个尺度较大的区域合并时,计算出的 T sl 并不太大,从而抑制错误合并的发生。

2.3 全局最优合并

2.3.1 基于RAG的全局最优合并
HAC局部最优合并结束后,整个图像被过分割,各个区域包含了一定数目的像元。相比于初始阶段,此时区域的数目减少了很多;另外,各个区域内的光谱异质性较大。为了提高合并的准确性,需要采用更为严谨的合并顺序和更为合理的合并规则。本文利用区域邻接图(Region Adjacency Graph,RAG)[23-24]来实现全局最优合并,以提高分割算法的精度。
RAG的实现包含2部分:节点 N M } (代表区域)和边 E = { e j j = 1,2 , , N E } (代表相邻区域之间的相似度),其中 N M N E 分别表示节点数和边数。由于RAG是无向图,任意2个相邻区域之间只存在一条边。一条边 e j = { m j 1 , m j 2 , w j 1 , j 2 } 包括连接的2个节点及表示它们相似程度的权值 w 。在合并过程中,对RAG的操作步骤为:
(1)初始化时,先设置参数,然后构建RAG;
(2)找到RAG中权值最小的边;
(3)若 e min 的权值 w min 小于阈值 T ,则先合并 e min 中的2个节点,再更新合并操作后的 M E ,最后返回步骤(2);否则,终止算法,输出结果。
其中,步骤(3)中的更新合并操作后的 M E 需要较多计算。本文采用了一种自平衡的二叉搜索树——红黑树(red-black tree)[25],将边 E 按其权值作为关键字进行排列,以提高更新 E 时边权值的搜索效率。红黑树将叶节点赋予红或黑的颜色属性。通过在插入或删除时对不同颜色节点位置的变换可使红黑树大致平衡,这保证了其搜索复杂度总是 O ( logn ) ,而普通二叉搜索树对于完全有序节点的搜索复杂度退化为 O ( n ) 图2显示了基于红黑树实现的全局最优合并算法,其中红黑树节点包含了3个变量:2个相邻区域的指针,以及这2个区域的相似度(式(5))。
Δ S dif ( i , j ) = 1 B b B ( μ bi - μ bj ) 2 (5)
式中: μ bi 表示区域 i 在波段 b 的光谱均值。
图2所示,构建red-black tree需要遍历RAG,并对相邻区域组成的节点进行插入操作。每完成一次合并,都需要把合并的节点从red-black tree中删除。不论是插入还是删除,都需要对红黑树进行搜索。由于红黑树的搜索复杂度为 O ( logn ) ,远低于一般数据结构的搜索复杂度 O ( n ) ,因此本文方法有效地提高了全局最优合并的计算效率。
Fig.2 The flowchart of global best merge based on red-black tree

图2 基于红黑树的全局最优合并算法流程图

需要注意的是,边权值的计算非常关键。不同算法对区域相似度的定义不同,从而导致算法性能的差异。本文的主要贡献在于对合并规则的优化。
2.3.2 C global 的设计
与局部最优合并不同,全局最优合并阶段中各个区域具有一定的大小,此时按照式(2)计算的光谱异质性差异值难以保证合并的正确性,因为区域大小(区域所包含的像元数目)对式(2)计算结果的影响可能大于光谱信息(区域中所有像元光谱值的标准偏差),从而导致错误的合并发生。因此,在构建RAG时,边权值(式(5))的计算仅考虑了区域的光谱信息。
在全局最优合并阶段,由于各个区域要通过合并来逼近真实地物,故本文在 C global 中放弃了形状信息。HR遥感影像中真实地物的形状差异巨大,导致紧凑异质性的数值变化远超过式(2)或式(5)的结果。图3显示了S1分割过程中光谱与紧凑异质性变化量随合并次数的变化。由此可知,在合并阶段后期,紧凑异质性的变化过大,易导致错误。
Fig.3 The variation of spectral and compactness heterogeneity with the increase of the merge number

图3 光谱与紧凑异质性随合并次数增长的变化

C global 还考虑了区域之间的边界信息,以防光谱相近、但边界明显的区域被错误地合并。本文采用的边界强度提取方法参考文献[23],其步骤为:
(1)利用高斯差分滤波方法计算各个波段的水平、竖直方向的梯度,构成 B × 2 维矩阵。
G x , y = v g 1 v g B h g 1 h g B (6)
式中: v g 1 表示波段1的像元 ( x , y ) 在竖直方向的灰度梯度值。
(2)像元的边界强度是矩阵GTG的最大特征值,2个区域的边界差异值是其公共边的边界强度的平均值。在全局最优合并过程中,每次合并前,都要考察待合并区域之间的边界强度,若其阈值大于 T edge ,则合并被禁止。 T edge 的计算公式如式(7)所示。
T edge = μ edge + λ σ edge (7)
式中: μ edge σ edge 分别是边界强度图像的灰度均值和标准偏差。 λ 是调节系数,一般取值为3.0。
在全局最优合并中,阈值 T sg 控制了合并数目的多少。 T sg 越大,合并次数越多,算法计算时间越短,但错误合并的概率越高,因此 T sg 的设置需要因影像而异。经多次实验发现,对于光谱范围为[0,255]的HR遥感影像,可将该阈值设置为10-30,以避免光谱差异较大的区域被合并。

3 实例验证与分析

3.1 实验数据

本文利用OrbView3获取的HR遥感数据进行算法验证。OrbView3是一颗商业遥感卫星,于2003年6月升空,可提供空间分辨率1 m的全色影像和4 m的多光谱影像。多光谱影像包含近红外、红色、绿色和蓝色4个波段。美国国家地质调查协会(United States Geological Survey,USGS)的官网提供了部分OrbView3数据的免费下载。
为了充分验证本文方法,分别采用一景较为简单的农田区域和一景较为复杂的城市区域的影像进行了分割实验,如图4所示。为了简便,下文对图4(a)、(b)分别简称为S1、S2。S1、S2的获取日期和中心经纬度分别为:2006年7月15日(38°11′45″N,121°25′36″W),2006年1月6日(36°05′22″N,120°26′36″E)。S1显示了美国加利福尼亚州的农田区域,其中,各个农田内部存在较大程度的光谱变化,且各个田地的大小具有较大差异,给影像分割带来了困难;S2显示了青岛市五四广场附近的区域。相比于S1,S2显然要复杂得多,很多房屋及其阴影构成了较小的区域,极易造成亚分割错误。
Fig.4 Two scenes of the OrbView3 multispectral images adopted by the experiment used in this paper

图4 本文实验所采用的2景OrbView3多光谱影像

3.2 评价方法与参数设置

为定量验证本文算法的精度,采用Crevier[27]提出的图像分割精度评价方法。该方法是一种监督评价方法,需要事先给出基准分割结果。为了简便,下文对算法的分割结果简称为S,对基准分割结果简称为G。图4(c)、(d)分别为实验所用2景数据的分割评价基准图。值得一提的是,由于S1较为简单,故所有的区域都被提取出来(图4(c));而S2较为复杂,将其所有区域都提取出来非常费时费力,故提取了其中具有代表性的区域(图4(d)中红色和蓝色部分),其中图4(d)中黑色部分未参与评价计算。
在评价过程中,需要计算准确率(Precision,P)和查全率(Recall,R)。P的定义为S和G中相匹配区域的交集与S中各个区域面积的比值(式(8))。
P = i = 1 n S S i S S ih G A (8)
式中: n S S 中区域的数目; S i S 表示S中的某个区域; S th G 表示在G中与 S i S 相匹配的区域; A G 中所有区域的面积和;| |表示集合的基数,即区域所包含的像元数目。R的定义为S和G中相匹配区域的交集与G中各个区域面积的比值(式(9))。
R = i = 1 n S S i S S th G S i S S th G S th G (9)
在计算P和R的过程中,要对S中的每个区域匹配G中的区域,其匹配标准为式(10)。
h = arg max j S i S S j G S i S S j G (10)
对于S中的某个区域 i ,当它与G中一个区域 j 的交集和并集的比值最大时,该区域就是 i 所匹配到的区域 h
P与R的数值范围是[0,1]。对于严重的过分割,P接近1而R接近0;相反,对于亚分割,P接近0而R接近1。只有当P和R同时都接近1时,才说明S与G非常相近。为了综合考虑过分割和亚分割对精度评价的影响,引入了F值(式(11))。
F = ( β 2 + 1 ) P R β 2 P + R (11)
式中: β 为调节系数,本文取值为2。F值越接近1,说明算法的分割精度越高。
本文算法需要设置2个参数,分别为式(7)的 λ 和全局最优合并的阈值 T sg 。经多次试验,λ和 T sg 对于S1的最佳参数,分别为3.0和20;对于S2,分别为3.0和12。RHSeg需要设置尺度参数,经过多次实验,S1和S2的参数分别为4000和800。MRS需要给出形状、紧凑性和尺度参数,其中前2个参数都被分别设置为0.1和0.5;S1和S2的尺度参数分别为110和50。

3.3 结果与分析

为了进行精度对比,本文利用RHSeg[11]和eCognition Developer Trial 8.0中的多分辨率分割算法(Multi-Resolution Segmenation,MRS)[26]与本文方法进行了对比,得到定量的分割精度评价结果。
3.3.1 农田地区影像的分割实验
图5显示了3种算法对S1的分割结果。本文方法很好地将S1中各个不同大小的农田区域分割出来,而RHSeg和MRS的分割结果中均存在过分割错误;虽然提高尺度参数会减少过分割,但2种算法若采用更大的尺度参数,会导致农田区域与周围细长的道路区域合并,这些错误的合并又会导致相邻的农田区域被合并,使分割精度显著降低。这主要是因为2种算法在分割过程中自始至终都采用了固定的合并标准函数,并且在函数中都考虑了区域大小信息。随着合并的进行,区域大小信息对合并准则函数的影响会越来越大,以致超过了光谱信息的影响,从而导致错误的合并产生。
Fig.5 Segmentation results using the 3 algorithms for S1

图5 3种算法的S1分割结果

由于本文算法采用了优化的合并规则,在合并过程中充分考虑了几何、光谱和边界信息,使分割精度得到显著地提高。表1列出了3种算法的P、R、F值,本文算法的F值明显高于另外2种算法,RHSeg的精度略好于MRS。与其他2种算法不同,本文算法在光谱异质性较大的区域容易产生细碎区域,如图5(b)东侧的农田。这与 C global 中未考虑区域大小信息有关,但这些细碎的区域对整体分割精度的影响并不显著。另外,表1还列出了3种算法的运算时间。本文方法的时间最短,这主要归功于先局部、后全局的合并策略;此外,利用红黑树实现全局最优合并也显著提升了运算速度。需要说明的是,本文实验均是在同一台笔记本电脑上完成的,其配置为:CPU是Intel Core I5 4200 m(2.5 GHz),内存4 GB,操作系统为Windows 7。
Tab.1 Quantitative evaluation of the 3 algorithms for S1

表1 3种算法的S1定量评价

P R F 时间/s
本文方法 0.9512 0.8624 0.8788 0.52
RHSeg 0.8540 0.7979 0.8085 0.66
MRS 0.9319 0.7450 0.7761 0.82

3.3.2 城市地区影像的分割实验

图6展现了S2的分割结果。与S1截然不同的是,S2包含了更多细小的区域,这极大地增加了分割的复杂度。本文算法的 T sg 比S1的小一些,目的是为了使阴影和道路等光谱十分相近的区域能够被区分。本文方法较为完整地保留了海水区域,同时很多楼房、草坪等细小区域也被较为完整地分割出来。表2中本文方法的F值最高,而MRS的亚分割错误最明显,其R值最高也印证了这一点。由于城区包含地物多,RHSeg在搜索合并的过程中耗时多,导致其运算时间最长。相反,本文方法在全局最优合并阶段采用了红黑树来提高搜索效率,明显降低了运算时间。
Tab.2 Quantitative evaluation of the 3 algorithms for S2

表2 3种算法的S2定量评价

P R F 时间/s
本文方法 0.8037 0.8758 0.8603 0.44
RHSeg 0.8674 0.7035 0.7311 0.96
MRS 0.2498 0.9183 0.5981 0.80
3.3.3 复杂影像的分割实验
为了进一步验证本文方法的性能,本文又采用5景尺度更大、复杂度更高的OrbView3影像开展分割实验。由于篇幅有限,图7仅展现了本文算法对一景较有代表性的影像的分割结果。该影像包含了农田和城市区域,且尺度更大(1000像元×1000像元),因此分割难度更高。图7(d)清晰地展示了该图东部区域放大的分割结果,一些细小的道路、农田、建筑物等地物都被较好地分割出来。 T sg 被设置为较小的值10,避免了光谱相近的区域被错误地合并,但也导致了较多细小的斑块。因此,在实际应用中,建议用户根据实际情况来调节 T sg
Fig.6 Segmentation results using the 3 algorithms for S2

图6 3种算法的S2分割结果

对于图7,本文方法的运算时间为2.90 s,显著低于MRS的4.43 s,说明随着影像尺度的增大,本文方法优越的合并策略与数据结构会增加算法运算速度的优势。5景影像分割结果的定量评价与S1和S2接近,进一步说明了本文方法的优越性。
Fig.7 The segmentation of a scene with relatively good representativeness

图7 一景较有代表性的遥感影像的分割结果

4 结论

本文提出了一种优化合并规则的HR遥感影像分割算法。该算法分为2个阶段,分别是基于HAC的局部最优合并和基于RAG的全局最优合并。在2个阶段中,综合考虑了光谱、几何与边界信息,并分别对合并规则进行了优化,以提高算法性能。在全局合并过程中,利用红黑树提高了搜索效率。通过对OrbView3中2景农田和城市子影像的分割实验,说明本文算法的分割精度显著优于RHSeg和MRS。本文主要结论如下:
(1)在基于合并的遥感影像分割算法中,应该综合考虑光谱、几何与边界信息,以获取最佳分割效果;
(2)在算法实现过程中,需要根据合并进行的阶段来定义不同的合并规则函数,以便提高算法的精度;
(3)在全局最优合并中,利用红黑树来提高RAG的搜索效率,可以加快算法的速度。
在今后的研究中,可利用其它类型的遥感器数据来进一步验证本文算法的性能,如SAR、Landsat或机载遥感影像等。另外,也有必要对本文算法的应用进行探究,如本文引言提到的GEOBIA、GIS矢栅数据转化和遥感影像压缩等。

The authors have declared that no competing interests exist.

[1]
Blaschke ., Hay ., Kelly ., et al.Geographic object-based image analysis - towards a new paradigm[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2014,87(1):180-191.The amount of scientific literature on (Geographic) Object-based Image Analysis – GEOBIA has been and still is sharply increasing. These approaches to analysing imagery have antecedents in earlier research on image segmentation and use GIS-like spatial analysis within classification and feature extraction approaches. This article investigates these development and its implications and asks whether or not this is a new paradigm in remote sensing and Geographic Information Science (GIScience). We first discuss several limitations of prevailing per-pixel methods when applied to high resolution images. Then we explore the paradigm concept developed by Kuhn (1962) and discuss whether GEOBIA can be regarded as a paradigm according to this definition. We crystallize core concepts of GEOBIA, including the role of objects, of ontologies and the multiplicity of scales and we discuss how these conceptual developments support important methods in remote sensing such as change detection and accuracy assessment. The ramifications of the different theoretical foundations between the ‘ per-pixel paradigm ’ and GEOBIA are analysed, as are some of the challenges along this path from pixels, to objects, to geo-intelligence. Based on several paradigm indications as defined by Kuhn and based on an analysis of peer-reviewed scientific literature we conclude that GEOBIA is a new and evolving paradigm.

DOI PMID

[2]
李苓苓,朱文泉,潘耀忠,等.基于高精度历史耕地地块的农区多光谱影像分割方法研究[J].国土资源遥感, 2011,23(4):20-25.<p>与传统的基于像元的影像分类方法相比,面向对象的分类方法能够提供更为准确的地类识别结果。对象作为信息提取过程的中间实体,对其划分的好坏直接关系到影像的分类精度。为了更准确地对农区多光谱影像进行分类,提出了一种基于高精度历史耕地地块数据的影像分割方法。该方法首先判定现势遥感图像上耕地地块的均质性,然后通过计算区域对比度指标指导局部最优参数的动态选择,最终获得影像的分割结果。基于农区的案例研究表明: 1在相同的全局分割参数条件下,基于高精度历史耕地地块数据的影像分割方法能够在保持稳定的耕地地块边界的同时,获得比直接影像分割更均一的对象; 2局部参数的应用使得影像&quot;欠分割&quot;和&quot;过分割&quot;现象基本消除,分割结果较使用全局参数的分割结果更为合理; 3局部最优分割参数的自动选取,极大地增强了该方法的客观性; 4区域对比度指标对&quot;欠分割&quot;和&quot;过分割&quot;现象十分敏感,能够较好地指导影像分割,为评价农区影像分割的优劣提供了有效的评价指标。</p>

DOI

[ Li L ., Zhu W ., Pan Y ., et al.Research on multi-spectral image segmentation of agriculture area based on high precision historical cropland parcels[J]. Remote Sensing for Land & Resources, 2011,23(4):20-25. ]

[3]
王琰,舒宁,龚龑,等.基于类别光谱变化规律的土地利用变化检测[J].国土资源遥感,2012,24(3):92-96.提出了一种基于类别光谱变化规律的高分辨率遥感图像土地利用变化检测方法。在基准期土地利用图的辅助下,以像斑为图像分析的基本单位,分别建立不同类别像斑特征在基准期和检测期图像上的分布曲线,通过三次多项式拟合参数表征上述2个时期特征值分布曲线的变化规律,在此基础上获取变化阈值,进行迭代计算,找出不符合类别光谱变化规律的像斑,确认为发生变化的像斑。以武汉市局部2002年、2005年QuickBird多光谱图像及相同区域2002年土地利用图为实验数据,以绿地和城区为例,对上述方法进行验证,证明上述方法有效。

DOI

[ Wang ., Shu ., Gong ., et al.Land use change detection based on class spectral change rule[J]. Remote Sensing for Land & Resources, 2012,24(3):92-96. ]

[4]
龚剑明,朱国强,杨娟,等.面向对象的南海珊瑚礁地貌单元提取[J].地球信息科学学报,2014,16(6):997-1004.<p>南海珊瑚礁地貌信息的提取是珊瑚礁资源利用、生态保护与管理及可持续发展的重要地学基础。本文提出了面向对象的珊瑚礁地貌单元提取模型, 针对不同的地貌单元, 以不同地貌单元的最优分割尺度、光谱参数、形状参数来分割影像并合并成不同对象, 从而获得相应的地貌单元。通过大量实验得出自然地貌的最优分割尺度区间为[140,600], 其中附礁生物稀疏带及丛生带、礁坑发育带的光谱参数和形状参数分别为0.9 和0.1, 其他自然地貌单元的光谱参数和形状参数分别为0.8 和0.2; 人工地貌的最优分割尺度区间为[25,170], 其光谱参数和形状参数分别为0.8 和0.2。进一步以南沙群岛簸箕礁WorldView-2 高分辨率遥感影像为例提取地貌单元, 并结合混淆矩阵和Kappa 系数对分类结果进行了精度评价, 地貌单元提取总体精度达到了85.75%, Kappa 系数为0.8349。结果表明, 该方法可有效运用南海珊瑚礁遥感影像的光谱特征、纹理特征, 以及影像数据不同波段的组合特性, 综合了影像和珊瑚礁地貌的关联特性, 充分利用了珊瑚礁不同地貌相带的异质性, 获得了理想的南海珊瑚礁地貌的整体信息, 满足了我国南海珊瑚礁地貌信息提取和地貌数字产品生成的需求。</p>

DOI

[ Gong J ., Zhu G ., Juan ., et al.A study on the object-oriented model for geomorphic unit extraction of coral reefs in the South China Sea[J]. Journal of Geoinformation Science, 2014,16(6):997-1004. ]

[5]
许夙晖,慕晓冬,柯冰,等.基于遥感影像的军事阵地动态监测技术研究[J].遥感技术与应用,2014,29(3):511-516.针对部队快速机动作战的军事要求,提出基于高分辨率遥感影像的军用阵地动态监测方法。借助面向对象的多尺度分割技术将阵地影像分割为同质对象,以提取各个对象的特征;针对监督分类和非监督分类的弊端,提出通过一定的先验知识制定分类规则的方法对遥感影像进行地物识别,在此基础上定性和定量地输出变化检测结果。实验结果表明:利用基于对象影像分析方法具有较高的识别精度,能够有效监测军事阵地变化。

DOI

[ Xu S ., Mu X ., Ke ., et al.A study on military battle-field surveillance technique based on remote sensing imagery[J]. Remote Sensing Technology and Application, 2014,29(3):511-516. ]

[6]
Elia ., Poggi ., Scarpa G.A tree-structured Markov random field model for Bayesian image segmentation[J]. IEEE Transactions on Image Processing, 2003,12(10):1259-1273.We present a new image segmentation algorithm based on a tree-structured binary MRF model. The image is recursively segmented in smaller and smaller regions until a stopping condition, local to each region, is met. Each elementary binary segmentation is obtained as the solution of a MAP estimation problem, with the region prior modeled as an MRF. Since only binary fields are used, and thanks to the tree structure, the algorithm is quite fast, and allows one to address the cluster validation problem in a seamless way. In addition, all field parameters are estimated locally, allowing for some spatial adaptivity. To improve segmentation accuracy, a split-and-merge procedure is also developed and a spatially adaptive MRF model is used. Numerical experiments on multispectral images show that the proposed algorithm is much faster than a similar reference algorithm based on "flat" MRF models, and its performance, in terms of segmentation accuracy and map smoothness, is comparable or even superior.

DOI PMID

[7]
Wu ., An J.An active contour model based on texture distribution for extracting inhomogeneous insulators from aerial images[J]. IEEE Transactions on Geoscience and Remote Sensing, 2014,52(6):3613-3626.The objects in natural images are often texturally inhomogeneous and prone to be falsely segmented into different parts by conventional methods. To overcome the difficulties caused by texture inhomogeneity, a new active contour model is proposed to extract inhomogeneous insulators from aerial images. First, a semilocal operator is employed to extract the texture features of insulators under the Beltrami framework. The layer of semilocal texture feature is single, and thus, it can avoid the high dimensionality of feature space. Then, a new convex energy functional is defined by taking the Xie's nonconvex model into a global minimization active contour framework during the process of segmentation. The proposed energy functional consists of not only the semilocal texture features of insulators but also their spatial relationship, which improves its ability to deal with textural inhomogeneity. Moreover, it can also avoid the existence of local minima in the minimization of the Xie's nonconvex model, thereby being independent of initial contour. In the process of contour evolution and numerical minimization, a fast dual formulation is employed to overcome the drawbacks of the usual level set and gradient descent method and to make the evolution of the contour more efficient. The experimental results on aerial insulator images confirm the ability of the proposed algorithm to effectively segment inhomogeneous textures with an overall average rmse of 1.87 pixels, a precision of 85.59%, and a recall of 86.47%. In addition, the proposed algorithm is extended to animal images, and satisfactory segmentation results can be obtained as well.

DOI

[8]
Bosworth ., Koshimizu ., Acton S.Multi-resolution segmentation of soil moisture imagery by watershed pyramids with region merging[J]. International Journal of Remote Sensing, 2013,24(4):741-760.A new image segmentation method is presented for application to remotely sensed imagery. The method incorporates the advantages of previous watershed segmentation techniques while greatly improving the two main disadvantages of such techniques: computational efficiency and control over region scale. Computational efficiency is attained through the use of a multi-resolution image pyramid, and control over region scale is accomplished through a variational region merging algorithm. We apply the segmentation method to remotely sensed estimates of surface soil moisture. Experimental results of computational efficiency and comparison with other segmentation methods are presented.

DOI

[9]
Beaulieu ., Goldberg M.Hierarchical in picture segmentation: a stepwise optimization approach[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1989,11(2):150-163.

[10]
Kurita T.An efficient agglomerative clustering algorithm for region growing[C]. MVA'94 IAPR Workshop on Machine Vision Applications, 1994:13-15.

[11]
Tilton ., Tarabalka ., Montesano ., et al.Best merge region-growing segmentation with integrated nonadjacent region object aggregation[J]. IEEE Transactions on Geoscience and Remote Sensing, 2012,50(11):4454-4467.Not Available

DOI

[12]
Baatz ., Schäpe A.Multiresolution segmentation: an optimizing approach for high quality multi-scale segmentation[A]. In: Angewandte Geographich Informationsverarbeitung XII[M]. Germany: Wichmann, 2000,12-23.

[13]
Vieira ., Formaggio ., Rennó ., et al.Object based image analysis and data mining applied to a remotely sensed landsat time-series to map sugarcane over large areas[J]. Remote Sensing of Environment, 2012,123(1):553-562.

[14]
Duro ., Franklin ., Dubé M.A comparison of pixel-based and object-based image analysis with selected machine learning algorithms for the classification of agricultural landscapes using Spot-5 HRG imagery[J]. Remote Sensing of Environment, 2012,118(1):259-272.

[15]
朱俊杰,杜小平,范湘涛,等.一种改进的多尺度分形网络演化分割方法[J].遥感技术与应用,2014,29(2):324-329.<p>面向对象技术是提高高分辨率图像分类精度的关键技术之一。针对eCognition的分形网络演化分割算法仅仅采用光谱特征和形状特征进行图像多尺度分割的不足,提出了将边缘特征引入其分割过程中,以提高多尺度分割的效果,来获得边缘平滑且分割对象与实际地物更加一致的分割结果,从而改善eCognition的分形网络演化分割方法中存在的过分割、欠分割和边缘粗糙等情况。通过实验,证明了引入边缘特征的分形网络演化分割方法提高了图像分割结果,减小了过分割和欠分割的产生,使得分割结果与实际地物更加一致。</p>

DOI

[ Zhu J ., Du X ., Fan X ., et al.An advanced multi-scale fractal net evolution approach for image segmentation[J]. Remote Sensing Technology and Application, 2014,29(2):324-329. ]

[16]
贾春阳,李卫华,李小春.基于自适应权值FNEA算法的高分辨率遥感影像分割[J].国土资源遥感,2013,25(4):22-25.<p>在高分辨率遥感图像分割方法中,分形网络演化算法(fractal net evolution approach,FNEA)是一种经典的影像对象构造方法。但在计算影像对象之间的异质性时,使用根据经验选择的固定权值会导致该算法不能很好地适应不同属性的影像对象分割。针对这一问题,提出了一种改进的FNEA方法,根据不同影像对象的空间和光谱特征,自适应地计算空间判据权值和紧凑度判据权值,并将不同光谱分量对光谱判据的贡献引入到影像对象之间异质性的计算中。计算机仿真实验结果表明,该文提出的算法对不同属性的影像对象具有很好的适应性,与同类算法相比,图像分割结果得到了较好的改善。</p>

DOI

[ Jia Y ., Li W ., Li X C.High resolution remote sensing image segmentation based on weight adaptive fractal net evolution approach[J]. Remote Sensing for Land & Resources, 2013,25(4):22-25. ]

[17]
Zhong ., Zhao ., Zhang L.Multiagent object-based classifier for high spatial resolution imagery[J]. IEEE Transactions on Geoscience and Remote Sensing, 2014,52(2):841-857.Not Available

DOI

[18]
Zhang ., Xiao ., Song ., et al.Boundary-constrained multi-scale segmentation method for remote sensing images[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2013,78(1):15-25.Image segmentation is the key step of Object-Based Image Analysis (OBIA) in remote sensing. This paper proposes a Boundary-Constrained Multi-Scale Segmentation (BCMS) method. Firstly, adjacent pixels are aggregated to generate initial segmentation according to the local best region growing strategy. Then, the Region Adjacency Graph (RAG) is built based on initial segmentation. Finally, the local mutual best region merging strategy is applied on RAG to produce multi-scale segmentation results. During the region merging process, a Step-Wise Scale Parameter (SWSP) strategy is proposed to produce boundary-constrained multi-scale segmentation results. Moreover, in order to improve the accuracy of object boundaries, the property of edge strength is introduced as a merging criterion. A set of high spatial resolution remote sensing images is used in the experiment, e.g., QuickBird, WorldView, and aerial image, to evaluate the effectiveness of the proposed method. The segmentation results of BCMS are compared with those of the commercial image analysis software eCognition. The experiment shows that BCMS can produce nested multi-scale segmentations with accurate and smooth boundaries, which proves the robustness of the proposed method.

DOI

[19]
Zhang ., Xiao ., Feng ., et al.Hybrid region merging method for segmentation of high-resolution remote sensing images[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2014,98(1):19-28.Image segmentation remains a challenging problem for object-based image analysis. In this paper, a hybrid region merging (HRM) method is proposed to segment high-resolution remote sensing images. HRM integrates the advantages of global-oriented and local-oriented region merging strategies into a unified framework. The globally most-similar pair of regions is used to determine the starting point of a growing region, which provides an elegant way to avoid the problem of starting point assignment and to enhance the optimization ability for local-oriented region merging. During the region growing procedure, the merging iterations are constrained within the local vicinity, so that the segmentation is accelerated and can reflect the local context, as compared with the global-oriented method. A set of high-resolution remote sensing images is used to test the effectiveness of the HRM method, and three region-based remote sensing image segmentation methods are adopted for comparison, including the hierarchical stepwise optimization (HSWO) method, the local-mutual best region merging (LMM) method, and the multiresolution segmentation (MRS) method embedded in eCognition Developer software. Both the supervised evaluation and visual assessment show that HRM performs better than HSWO and LMM by combining both their advantages. The segmentation results of HRM and MRS are visually comparable, but HRM can describe objects as single regions better than MRS, and the supervised and unsupervised evaluation results further prove the superiority of HRM.

DOI

[20]
苏腾飞,孟俊敏,张晰.基于HAC的溢油SAR影像分割算法[J].海洋科学进展,2013,31(2):256-265.图像分割是SAR溢油检测中的 关键步骤,但由于SAR影像中存在斑点噪声,使得一般的图像分割算法难以收到理想的效果,严重影响溢油检测的精度。发展一种基于凝聚层次聚类 (Hierarchical Agglomerative Cluste-ring,HAC)的溢油SAR图像分割算法。该算法利用多尺度分割的思想,能够有效保持SAR影像中溢油斑块的形状特征,并能减少细碎斑 块的产生。利用2010年墨西哥湾的Envisat ASAR影像开展了溢油SAR图像分割实验,并将该算法和Canny边缘检测、OTSU阈值分割、FCM分割、水平集分割等方法进行了对比。结果显 示,HAC方法可以有效减少细碎斑块的产生,有助于提高SAR溢油检测的精度。

DOI

[ Su T ., Meng J ., Zhang X.A segmentation algorithm for oil spill SAR image based on hierarchical agglomerative clustering[J]. Advances in Marine Science, 2013,31(2):256-265. ]

[21]
Dabboora ., Yackel ., Hossain ., et al.Comparing matrix distance measures for unsupervised POLSAR data classification of sea ice based on agglomerative clustering[J]. International Journal of Remote Sensing, 2013,34(4):1492-1505.

[22]
Moran P A .. Notes on continuous stochastic phenomena[J]. Biometrika, 1950,37(1):17-23.

PMID

[23]
Qin ., Clausi D.Multivariate image segmentation using semantic region growing with adaptive edge penalty[J]. IEEE Transactions on Image Processing, 2010,19(8):2157-2170.Not Available

DOI PMID

[24]
Felzenszwalb P ., Huttenlocher D P.Efficient graph-based image segmentation[J]. International Journal of Computer Vision, 2004,59(2):167-181.<a name="Abs1"></a>This paper addresses the problem of segmenting an image into regions. We define a predicate for measuring the evidence for a boundary between two regions using a graph-based representation of the image. We then develop an efficient segmentation algorithm based on this predicate, and show that although this algorithm makes greedy decisions it produces segmentations that satisfy global properties. We apply the algorithm to image segmentation using two different kinds of local neighborhoods in constructing the graph, and illustrate the results with both real and synthetic images. The algorithm runs in time nearly linear in the number of graph edges and is also fast in practice. An important characteristic of the method is its ability to preserve detail in low-variability image regions while ignoring detail in high-variability regions.

DOI

[25]
Cormen T ., Charles E ., Rivest R ., et al.Introduction to algorithms, second edition[M]. Massachusetts: The MIT Press, 2001:238-239.

[26]
Trimble. eCognition developer 8 reference book[EB/OL]. .

[27]
Crevier D.Image segmentation algorithm development using ground truth image data sets[J]. Computer Visual and Image Understanding, 2008,112(2):143-159.A methodology is presented for making use of ground truth, human-segmented image data sets to compare, develop and optimize image segmentation algorithms. Central to this question is the problem of quantifying the accuracy of the match between machine and reference segmentations. In this regard, the paper introduces a natural extension to the concept of precision-recall curves, which are a standard evaluation technique in pattern recognition. Computationally efficient match measures defined so as to benefit from the availability of multiple alternative human segmentations, are also proposed. The Berkeley image segmentation data set is used to select among the proposed measures, which results in a validation of the local best fit heuristic as a way to best exploit reference segmentations. I then show how the resulting match criterion can be used to improve the recent SRM segmentation algorithm by gradual modifications and additions. In particular, I demonstrate and quantify performance increases resulting from changing color coordinates, optimizing the segment merging rule, introducing texture, and forcing segments to stop at edges. As modifications to the algorithm require the optimization of parameters, a mixed deterministic and Monte-Carlo method well adapted to the problem is introduced. A demonstration of how the method can be used to compare the performance of two algorithms is made, and its broad applicability to other segmentation methods is discussed.

DOI

Options
文章导航

/