融合像元形状和光谱信息的高分遥感图像分类新方法

引用本文

杨青山, 张华. 融合像元形状和光谱信息的高分遥感图像分类新方法[J]. 国土资源遥感, 2016,28(4): 64-70
YANG Qingshan, ZHANG Hua. A new method for classification of high spatial resolution remotely sensed imagery based on fusion of shape and spectral information of pixels[J]. REMOTE SENSING FOR LAND & RESOURCES,2016,28(4): 64-70 复制到剪切板

Permissions

《国土资源遥感》编辑部

融合像元形状和光谱信息的高分遥感图像分类新方法

杨青山¹, 张华²

1.武汉大学遥感信息工程学院,武汉 430079

2.中国矿业大学(徐州)环境与测绘学院,徐州 221116

张华(1979-),男,博士,副教授,主要从事遥感数据不确定性、空间分析及GIS算法与应用系统开发等方面的研究。Email:zhhua_79@163.com。

第一作者简介: 杨青山(1993-),男,硕士研究生,主要研究方向为GIS和遥感图像处理与特征提取。Email:wilm_yang@foxmail.com。

收稿日期: 2016-01-13

摘要

在高空间分辨率(简称“高分”)遥感图像分类中,由于存在“同谱异物”等现象,仅依靠光谱信息进行分类的误差较大。为提高图像分类精度,提出一种融合像元形状和光谱特征信息的高分多光谱遥感图像分类新方法。首先利用像元及其邻域的关系来描述其空间结构,计算并提取像元同质区域(pixel homogeneous regions,PHR); 然后以所提取的同质区域为基础,分别计算中心像元的长/宽比(length-width ratio,LW)和面积/周长比(area-perimeter ratio,PAI)这2个像元形状特征; 最后将归一化后的像元形状特征和光谱特征融合,并利用支持向量机分类方法进行分类。以2个区域的QuickBird高分遥感图像对该算法进行验证,将实验结果与仅利用光谱信息分类和仅使用像元形状指数(pixel shape index,PSI)分类的结果进行比较。结果表明,所提出的方法得到的分类精度最高,该方法能有效地提高高分遥感图像的分类精度。

关键词: 像元同质区域(PHR); 像元形状指数(PSI); 阈值; 高空间分辨率遥感图像

文献标志码:A 文章编号:1001-070X(2016)04-0064-07 doi: 10.6046/gtzyyg.2016.04.10

A new method for classification of high spatial resolution remotely sensed imagery based on fusion of shape and spectral information of pixels

YANG Qingshan¹, ZHANG Hua²

1. School of Remote Sensing and Information Engineering, Wuhan University, Wuhan 430079, China

2. School of Environment Science and Spatial Informatics, China University of Mining and Technology(Xuzhou), Xuzhou 221116, China

Abstract

In the classification of high spatial resolution remotely sensed imagery,due to the presence of the same object with different spectra, the dependence only on spectral information for classification is not enough. To improve the accuracy of classification, the authors proposed a novel spatial features extraction method for classification of the HSRMI. Firstly, neighborhood pixels’ spatial relationship was described and used to calculate and extract the pixel homogeneous regions (PHR). Then, based on the extracted PHR, the pixels’ shape index features, including length-width ratio(LW) and area-perimeter ratio(PAI), were extracted. Lastly, the pixel shape index features were normalized and combined with the spectral information to perform classification by using SVM classification method. Two different areas’ QuickBird images were used to test the performance of proposed method. The experimental results show that the proposed method has the highest performance compared with pixel shape index(PSI)and spectral information, and can improve the classification accuracy of high spatial resolution remotely sensed imagery.

Keyword: pixel homogeneous regions(PHR); pixel shape index(PSI); threshold; high spatial resolution remotely sensed imagery

Show Figures

0 引言

传统的基于像元的遥感数据监督分类算法主要根据地物光谱数据, 统计以像元为基本单元的图像在不同光谱数据组合中的差别, 对图像进行分类。但由于“ 同谱异物” 现象和大量混合像元的存在, 基于像元光谱特征的分类方法往往不能达到满意的分类效果。近年来国内外遥感研究人员提出了很多用于提高遥感图像分类精度的方法来弥补仅依靠光谱特征进行分类的不足。郑淑丹等^[1]基于分形理论和灰度共生矩阵(gray-level co-occurrence matrix, GLCM)2种方法提取不同的纹理特征, 结合光谱信息对遥感图像进行最大似然监督分类, 得到较高的分类精度。李厚强等^[2]将纹理特征和归一化的色度和饱和度输入反向传播(back prepagation, BP)神经网络分类器进行遥感图像分类, 较好地解决了光谱特征相似地物的区分问题。陈秋晓等^[3]从图像分割获得的影像中提取光谱特征和拓扑特征等信息, 使影像易于识别和分类。张锦水等^[4]基于支持向量机(support vector machine, SVM)的分类方法, 利用复合光谱、纹理和结构信息等多特征数据对IKONOS高空间分辨率(简称“ 高分” )图像进行分类, 取得较高的分类精度。黄昕等^[5]提出像元形状指数(pixel shape index, PSI), 利用像元邻域的光谱相似性探测其空间形状特征, 在高分遥感图像分类中具有一定的优势。Shackelford等^[6]提出一种长度-宽度提取算法, 用于提取光谱特性相似的一组像元的长度与宽度信息, 并得出该算法适合对不同长/宽比对象进行分类的结论。Han等^[7]提出的形状尺寸指数(shape-size index, SSI)则是基于频谱相似的同质区域中的中心像元及其相邻像元的特征信息提取的。国内外学者已进行的分类研究表明, 从遥感图像中可以提取大量的形状、纹理和空间关系等特征信息, 用于提高分类精度。

本文提出利用PSI, 根据像元及其相邻像元共同表征的形状特征来提取影像的形状特征信息; 再将归一化后的形状特征和光谱特征输入SVM进行高光谱遥感图像分类; 详细讨论了像元形状特征的提取方法以及长/宽比(length-width ratio, LW)和面积/周长比(area-perimeter ratio, PAI)以及融合二者的PL(PAI- LW)形状特征对分类结果的影响; 并将实验结果与仅利用光谱信息分类和仅使用PSI分类的结果进行比较。实验证明, 在分类算法中融入对像元形状特征的分析, 能够提高分类精度。

1 像元形状特征

定义相邻像元之间光谱相似的区域为像元同质区域(pixel homogeneous regions, PHR), 在PHR的基础上提取2种像元形状特征指标, 进行图像分类。

1.1 PHR确定

为了提高对空间特征的分类精度, 利用中心像元与其8个相邻像元之间的频谱相似性计算PHR。

1.1.1 确定边缘像元

边缘像元的计算公式为

edge(i)= $\begin{matrix} \frac{1}{N} \end{matrix} \begin{matrix} \overset{N}{\sum_{d = 1}} \end{matrix}$ edge_d(i) , (1)

式中: edge(i)为估计概率值, 表示第i个像元属于边缘像元的可能性; N为波段的个数; edge_d(i)为d波段的edge(i)值。edge(i)值用于判断d波段中的第i个像元是否是边缘像元: 当edge(i)=1时, 该像元是边缘像元; 当edge(i)=0时则不是。edge(i)的值用Canny滤波器获得。

1.1.2 计算光谱相似性

光谱相似性的计算公式为

PH(i)= $\begin{matrix} \underset{x \in X}{arg min} \end{matrix}$ [1+edge(i)] $\begin{matrix} \overset{N}{\sum_{d = 1}} \end{matrix}$ |P_d(i)-P_d(x)| , (2)

式中: PH(i)为第i个像元与其周围像元的光谱相似性; arg min为使目标函数取最小值时的变量值; P_d(i)和P_d(x)分别为d波段中的中心像元和与其相邻像元的光谱值; x为与中心像元相邻的8个像元。edge(i)值越大, 该像元在此同质区域中的可能性就越小。

1.1.3 形成中心像元的PHR

给定一个阈值T, 若PH(i)< T, 则可认为第i个像元包含在此中心像元所在的PHR内, 将被合并在PHR内; 这时, 中心像元的光谱值将被改变。新的中心像元的光谱值的计算公式为

$\begin{matrix} P_{d}^{n + 1} \end{matrix}$ (i)= $\begin{matrix} \frac{n P_{d}^{(n)} (i) + P_{d} (c)}{n + 1} \end{matrix}$ , (3)

式中: $\begin{matrix} P_{d}^{n + 1} \end{matrix}$ (i)为新的中心像元的光谱值; $\begin{matrix} P_{d}^{(n)} \end{matrix}$ (i)为合并之前PHR中的中心像元的光谱值; n为合并之前PHR中像元的个数; P_d(c)为新加入PHR的像元的光谱值。

重复进行光谱相似性计算和像元合并运算, 直到没有新的像元需要合并, 最终得到中心像元i的PHR(图1)。

	Figure Option View Download New Window
	图1 像元形状特征计算示意图Fig.1 Sketch map for computation of pixel shape feature

图1中红色像元是中心像元, 灰色像元是PHR中除中心像元和边缘像元之外的所有像元, 绿色像元是PHR的边缘像元。在所求出的PHR中进行指标提取的计算, 遍历图像的所有像元, 计算每1个像元的形状特征, 并进行分类。

1.2 阈值设定

在构造PHR的过程中, 用来与PH(i)进行比较的阈值通常都是人工设定的, 可能会造成PHR提取结果有较大误差。因此, 本文采用自适应阈值的方法设定阈值。在本质上, 图像的边缘像元可以表示整个图像的局部不连续性, 并可近似地表示整个图像的特征, 所以可利用PHR的边缘像元获得最佳阈值, 即

T(i)= $\begin{matrix} \overset{N}{\sum_{d = 1}} \end{matrix} \begin{matrix} |P_{d} (i) - \frac{1}{M_{d}} \overset{M_{d}}{\sum_{k = 1}} X_{d} (k)| \end{matrix}$ , (4)

式中: T(i)为第i个像元的PHR的阈值; M_d为波段d中的像元数目; X_d(k)为边缘像元在d波段中的光谱值。

1.3 像元形状特征提取

本文在充分考虑相邻像元的光谱相似性和相同形状区域内的像元具有相近特征值的情况下, 提出了2个像元PHR的形状特征值, 即长/宽比(LW)和面积/周长比(PAI)。

1.3.1 LW

LW的计算公式为

LW=P_Length/P_Width , (5)

式中: P_Length为与中心像元所在的PHR具有相同标准二阶中心矩的椭圆的长轴长度; P_Width为与中心像元所在的PHR具有相同标准二阶中心矩的椭圆的短轴长度(参见图1)。

1.3.2 PAI

PAI的计算公式为

PAI=P_Perimeter/P_Area , (6)

式中: P_Perimeter为中心像元所在PHR的边缘像元个数之和; P_Area为中心像元所在PHR内所有像元个数之和。当PHR面积相同时, PAI越大, 则区域越复杂。形状特征提取的优势在于: ①考虑了边缘像元; ②自适应阈值用于获得PHR; ③多边形的特征在不确定的窗口内。

2 实验方法与步骤

本文的实验方法是基于光谱和形状特征, 利用 SVM分类器进行分类: ①获得1景高分多光谱图像, 用Canny滤波器^[8]提取边缘信息, 得到模糊边缘图; ②基于Canny滤波器和原始图像, 用所设定的阈值计算每1个像元的PHR; ③基于模糊边缘图、自适应阈值、原始图像以及基于PHR的像元形状特征, 获得每1个像元的PHR; ④应用SVM分类器^[9]对图像进行基于光谱和形状特征的分类。SVM分类器的核函数为径向基函数(radial basis function, RBF), 采用交叉验证获取最优参数C和g的值^[9]。PL为综合了PAI和LW的分类结果。

实验步骤如下:

1)获取模糊边缘图。为了在求取原始图像的最小变化值和保存边缘信息时避免人工的干扰, 采用中值滤波器, 首先用一个3× 3邻域窗口对图像进行运算; 然后用 Canny算法获得滤波后图像中的边

缘信息, 得到edge(i)值; 最后用式(1)计算得到模糊边缘图。

2)按照前文提到的方法设定一个合适的阈值。

3)获取像元的PHR。利用获取的模糊边缘图, 根据设定的阈值选取像元的PHR。

4)获得像元的形状特征集。形状特征包括长/宽比(LW)和面积/周长比(PAI)。

5)基于SVM分类。将提取的像元形状特征和多光谱特征均标准化为[0, 1], 使所提取的形状特征和多光谱特征集成1个整体特征(PL); 然后应用SVM分类器对所集成的图像进行分类。

3 实验结果与分析

本文实验采用遥感图像和地面参考数据作为分类数据。分类算法是广泛使用的基于Matlab7.8、由LIBSVM提供的SVM库。用2个实验论证本文提出的新方法。将像元形状特征和光谱特征组合作为分类特征, 输入SVM分类器中进行分类。

3.1 实验一

3.1.1 研究数据

实验采用分层随机抽样的方法, 从地面参考数据中分别抽取各类别样本数据总数的1%作为训练样本数据集, 其余数据作为检验样本数据集。原始遥感图像、地面实况图像、实验用样本及提取的不同形状特征图像如图2所示。

	Figure Option View Download New Window
	图2 实验一样本及形状特征影像Fig.2 Samples and shape feature of image for experiment 1

3.1.2 分类结果分析

图2(e)(f)是提取的2种像元形状特征, 将其与图像的光谱信息作为分类的特征信息输入SVM分类器, 分别对不同的形状特征数据进行了分类, 结果见图3。

	Figure Option View Download New Window
	图3 实验一SVM分类结果Fig.3 Results of SVM classification for experiment 1

通过实验一可知, 阴影、草地和树木的光谱特性很相似, 故仅使用光谱信息很难得到好的分类效果。要想提高分类精度, 必须引入其他的分类特征信息。图3(a)是仅使用光谱信息分类的分类结果; 图3(b), (c), (d)和(e)分别是加入了PAI, LW, PL和PSI像元形状特征后的分类结果。从表1可以看出, 加入像元形状特征后的分类精度得到了大幅度的提高, 最高的(PL)分类总体精度比仅使用光谱信息分类提高了8.47%。不同的像元形状特征对各类地物分类精度的提高程度也不同, 从各个类的分类精度提高情况来看, 虽然单一利用PAI和LW形状特征对某些地物的分类精度比PSI的低, 但是综合了PAI和LW的PL形状特征对各个类的分类精度的提高最为明显, 最终的总体分类精度达到了94.66%, Kappa系数达到了0.932 1, 均比PSI的分类精度高。说明在分类图像中存在各种复杂形状的图斑, 很难用单一的形状特征加以区分和识别, 在实际分类中应采用多种像元形状特征组合来提高分类的精度。

表1 实验一的分类精度统计 Tab.1 Classification accuracy statistics for experiment 1

3.2 实验二

3.2.1 研究数据

实验二同样采用分层随机抽样的方法, 从地面参考数据中分别抽取了各类别样本数据总数的1%作为训练样本数据集, 其余数据作为验证样本数据集。实验二样本及不同形状特征图像如图4所示。

	Figure Option View Download New Window
	图4 实验二样本及形状特征影像Fig.4 Samples and shape feature images for experiment 2

3.2.2 分类结果分析

图4(e)(f)是提取的2种像元形状特征的图像, 将其和图像的光谱特征信息作为分类的特征信息输入SVM分类器, 分别对不同的形状特征数据进行了分类, 结果如图5所示。

	Figure Option View Download New Window
	图5 实验二 SVM分类结果Fig.5 Results of SVM classification for experiment 2

从实验二的分类效果来看, 各种加入像元形状特征后的分类效果要比仅使用光谱特征进行分类的效果好。图5(a)是仅使用光谱信息的分类结果, 图中水体与阴影、道路与房屋、裸地与房屋及草地与树的错分现象比较严重, 其原因是它们的光谱特性太相似, 故仅利用光谱信息来区分是很难达得好的分类效果的, 必须加入形状特征等其他信息。图5(b), (c), (d)和(e)分别是加入了PAI, LW, PL和PSI像元形状特征后的分类结果。从表2可以看出, 加入像元形状特征后的分类精度得到了大幅度的提高, 其中PL分类的总体分类精度达到了93.47%, 在所有特征组合分类精度中最高, 比仅利用光谱信息提高了18.61%, 对树木、裸地、道路、阴影、草地、房屋和水体分类精度分别提高了57.69%, 50.70%, 43.10%, 39.65%, 6.01%, 5.33%和4.19%。PL对各个类的分类精度的提高是所有分类特征中最大的, 再次说明了分类图像中存在各种形状复杂的图斑, 很难用单一的形状特征来识别, 在实际分类中应该采用多种像元形状特征组合来提高分类的精度。

表2 实验二的分类精度统计 Tab.2 Classification accuracy statistics for experiment 2

4 结论

1)由于地物之间的光谱相似性, 传统的仅利用光谱信息的分类方法不能得到精度满意的分类结果; 要想提高分类精度, 需要更多考虑图像中地物的结构、形态、分布等空间特征信息的遥感图像分类技术。

2)引入长/宽比(LW)和面积/周长比(PAI)这2个像元形状特征, 与影像的光谱信息组合作为SVM分类器的输入数据, 可提高遥感图像的分类精度。2个实验的结果表明: 所引入的像元形状特征充分考虑了相邻像元间关系等形状和空间位置特征, 能大幅度提高遥感图像的分类精度; 不同像元形状特征对不同遥感图像中不同地物的分类精度的提高程度也有所不同。

3)1种像元形状特征难以适应不同的地物类别, 所以在实际分类过程中, 应将多种形状特征组合用于遥感图像分类, 以得到相对满意的分类结果; 对于多特征组合分类, 一般需要进行特征选择和优化。

4)本文中所涉及的形状特征有限, 还需研究更多的特征, 以保证所得结论的普适性和通用性。在特征提取中, 对尺度的影响及在像元形状特征计算中的阈值选择规则等问题尚须进一步研究。

The authors have declared that no competing interests exist.

参考文献

View Option

[1]	郑淑丹, 郑江华, 石明辉, 等. 基于分形和灰度共生矩阵纹理特征的种植型药用植物遥感分类[J]. 遥感学报, 2014, 18(4): 868-886. Zheng S D, Zheng J H, Shi M H, et al. Classification of cultivated Chinese medicinal plants based on fractal theory and gray level co-occurrence matrix textures[J]. Journal of Remote Sensing, 2014, 18(4): 868-886. [本文引用:1]
[2]	李厚强, 刘政凯, 林峰. 基于分形理论的航空图像分类方法[J]. 遥感学报, 2001, 5(5): 353-357. Li H Q, Liu Z K, Lin F. Aerial image classification method based on fractal theory[J]. Journal of Remote Sensing, 2001, 5(5): 353-357. [本文引用:1]
[3]	陈秋晓, 骆剑承, 周成虎, 等. 基于多特征的遥感影像分类方法[J]. 遥感学报, 2004, 8(3): 239-245. Chen Q X, Luo J C, Zhou C H, et al. Classification of remotely sensed imagery using multi-features based approach[J]. Journal of Remote Sensing, 2004, 8(3): 239-245. [本文引用:1]
[4]	张锦水, 何春阳, 潘耀忠, 等. 基于SVM的多源信息复合的高空间分辨率遥感数据分类研究[J]. 遥感学报, 2006, 10(1): 49-57. Zhang J S, He C Y, Pan Y Z, et al. The high spatial resolution RS image classification based on SVM method with the multi-source data[J]. Journal of Remote Sensing, 2006, 10(1): 49-57. [本文引用:1]
[5]	黄昕, 张良培, 李平湘. 融合形状和光谱的高空间分辨率遥感影像分类[J]. 遥感学报, 2007, 11(2): 193-200. Huang X, Zhang L P, Li P X. Classification of high spatial resolution remotely sensed imagery based on the fusion of spectral and shape features[J]. Journal of Remote Sensing, 2007, 11(2): 193-200. [本文引用:1]
[6]	Shackelford A K, Davis C H. A hierarchical fuzzy classification approach for high-resolution multispectral data over urban areas[J]. IEEE Transactions on Geoscience and Remote Sensing, 2003, 41(9): 1920-1932. [本文引用:1]
[7]	Han Y, Kim H, Choi J, et al. A shape-size index extraction for classification of high resolution multispectral satellite images[J]. International Journal of Remote Sensing, 2012, 33(6): 1682-1700. [本文引用:1]
[8]	薛丽霞, 李涛, 王佐成. 一种自适应的Canny边缘检测算法[J]. 计算机应用研究, 2010, 27(9): 3588-3590. Xue L X, Li T, Wang Z C. Adaptive Canny edge detection algorithm[J]. Application Research of Computers, 2010, 27(9): 3588-3590. [本文引用:1]
[9]	付纪刚. 基于支持向量机的边缘检测算法研究[D]. 西安: 西安科技大学, 2011. Fu J G. Research of Edge Detection Based on Support Vector Machine[D]. Xi’an: Xi’an University of Science and Technology, 2011. [本文引用:2]
[10]	张华. 遥感数据可靠性分类方法研究[D]. 徐州: 中国矿业大学, 2012. Zhang H. Study on Reliable Classification Methods Based on Remotely Sensed Image[D]. Xuzhou: China University of Mining and Technology, 2012. [本文引用:1]
[11]	Chang C C, Lin C J. LIBSVM: A library for support vector machines[EB/OL]. 2001. http://www.csie.ntu.edu.tw/~cjlin/libsvm/. [本文引用:1]
[12]	Huang X, Zhang L P, Li P X. Classification and extraction of spatial features in urban areas using high-resolution multispectral imagery[J]. IEEE Geoscience and Remote Sensing Letters, 2007, 4(2): 260-264. [本文引用:1]
[13]	Zhang L P, Huang X, Huang B, et al. A pixel shape index coupled with spectral information for classification of high spatial resolution remotely sensed imagery[J]. IEEE Transactions on Geoscience and Remote Sensing, 2006, 44(10): 2950-2961. [本文引用:1]

2014

0.0

郑淑丹, 郑江华, 石明辉, 等. 基于分形和灰度共生矩阵纹理特征的种植型药用植物遥感分类[J]. 遥感学报, 2014, 18(4): 868-886.

Zheng S

, Zheng J

, Shi M

, et al. Classification of cultivated Chinese medicinal plants based on fractal theory and gray level co-occurrence matrix textures[J]. Journal of Remote Sensing, 2014, 18(4): 868-886.

Remote sensing has become an important exploratory method for the national census of traditional Chinese medicinal resources. We selected safflower (Latin name: Carthamus tinctorius L.) as a sample species of cultivated Chinese medicinal plant species. We extracted textures from ZY-3 satellite images based on fractal theory and Gray Level Co-occurrence Matrix (GLCM) methods to assist the supervised classification. Results showed that the overall accuracy increased by 0.49% to 5.31%. Kappa coefficient increased by 0.01 to 0.07 after textures were combined for classification processing. Accuracy with fractal-based textures improved at least twice as much as GLCM-based textures, particularly when fractal-based texture was extracted with a 5×5 sliding block in Matlab and then added. The classification accuracy of safflower increased to 100% when this parameter was combined with fractal-based textures; by contrast, this accuracy reduced by 0.55% to 1.28% with GLCM-based textures. Moreover, the samples collected from the final recognition results based on fractal theory were relatively complete with a smaller degree of fragmentation and a higher distinction from other categories. Therefore, fractal-based textures can be used to assist in the recognition of cultivated Chinese medicinal plants. Textures based on fractal theory could effectively increase the classification accuracy of ZY-3 images at a higher extent than those based on GLCM.

遥感技术应用已成为中国中药资源普查的一个重要探索方向。以红花（Carthams Tinctorius L.）为种植型药用植物实验样本品种，分别基于分形理论和灰度共生矩阵（GLCM）两种方法提取不同的纹理特征，结合光谱信息对资源三号卫星（ZY-3）影像进行最大似然方法的监督分类，对比分析分类效果和精度评价。结果显示：加入纹理特征后，总体分类精度提高了0.49%-5.31%，Kappa系数提高了0.01-0.07，结合基于双毯法的分形纹理较GLCM纹理分类总体效果提高至少两倍，其中在Matlab环境下，使用5×5滑动窗口提取的分形纹理特征的分类效果最显著，总体分类精度提高了5.31%，Kappa系数提高了0.07。对于红花分类精度，引入分形纹理特征的分类精度提高到了100%，识别的红花样地效果最完整，破碎程度最小，与其他类别区分度最高；而引入GLCM的分类精度却降低了0.55%-1.28%，可见采用分形理论比采用GLCM提取纹理特征能够更加有效地辅助ZY-3影像识别种植型药用植物。

... 郑淑丹等^[1]基于分形理论和灰度共生矩阵(gray-level co-occurrence matrix,GLCM)2种方法提取不同的纹理特征,结合光谱信息对遥感图像进行最大似然监督分类,得到较高的分类精度 ...

2001

0.0

李厚强, 刘政凯, 林峰. 基于分形理论的航空图像分类方法[J]. 遥感学报, 2001, 5(5): 353-357.

Li H

, Liu Z

, Lin

Aerial image classification method based on fractal theory[J]. Journal of Remote Sensing, 2001, 5(5): 353-357.

Remote sensing images have both spectral and textural features. How to make uses of these features is very important to the practical work of remote sensing image classification. This paper presents a supervised classification method of aerial remote sensing image, which takes advantages of both spectral features and textural features. First, this paper puts forward a set of textural features with their computation approaches based on fractal and multifractal theory, including fractal dimension, multifractal function %q_D(q)%, and lacunarity. The fractal_based textural features are relatively insensitive to the image scaling, therefore, within certain scope, the fractal_based textural features obtained from a remote sensing image under one resolution can also be used in the remote sensing images under other resolutions. This is very valuable in practice. Then, this paper presents the classification method which consists of two parts, namely feature extraction and classifier construction. In the part of feature extraction, this method converts color aerial image from RGB to HSI and computes fractal dimension, multifractal function %q_D(q)%, and lacunarity by intensity as texture features with normalized hue and saturation being used as spectral features. In the part of classifier construction, it adopts BP neuval network as classifier. In the end, the experiment of classifying the aerial images has been done and the result is satisfactory, which verifies the effect of this methood.

提出一种基于分形理论和BP神经网络的航空遥感图像有监督分类方法。该方法尝试将航空图像的光谱信息和纹理特征相结合。它首先将彩色航空图像由RGB格式转化为HSI格式，然后，根据亮度计算分数维、多重分形广义维数谱q-D(q)和“空隙”等基于分形的纹理特征，同时加入归一化的色度和饱和度作为光谱特征，采用BP神经网络作为分类器。通过对彩色航空图像的分类实验，结果证实该方法行之有效。

... 李厚强等^[2]将纹理特征和归一化的色度和饱和度输入反向传播(back prepagation,BP)神经网络分类器进行遥感图像分类,较好地解决了光谱特征相似地物的区分问题 ...

2004

0.0

陈秋晓, 骆剑承, 周成虎, 等. 基于多特征的遥感影像分类方法[J]. 遥感学报, 2004, 8(3): 239-245.

Chen Q

, Luo J

, Zhou C

, et al. Classification of remotely sensed imagery using multi-features based approach[J]. Journal of Remote Sensing, 2004, 8(3): 239-245.

In this paper, a new method, namely the multi feature based method, was presented to classify remotely sensed imagery. A class scheme was firstly defined, in which, every class was decomposed into subclasses if necessary. Then, some suitable features were selected, and every subclass was characterized with a unique combination of these features. Meanwhile, the whole imagery was segmented into image objects, and subsequently the features used were measured, like spectral, geometrical and topological features. Using these features, image objects could be recognized and classified easily. By using the multi features based classification approach, classification accuracy was improved, and the classification results were easier interpreted when compared with the conventional classification method.

提出了一种基于多特征的遥感分类方法。首先 ,制定类方案并分解各个类 ,据此得到相应的子类 ;然后 ,通过选用适当的特征 ,使得每一个类都能以一个独特的特征组合来表达。与此同时 ,通过影像分割得到影像对象 ,并测量这些对象的各个特征 ,如光谱特征、几何特征及拓扑特征等。凭借这些特征 ,影像对象可以较为容易地被识别和分类。与传统的分类方法的比较表明 ,文中所提出的分类方法具有明显的优越性和良好的前景。

... 陈秋晓等^[3]从图像分割获得的影像中提取光谱特征和拓扑特征等信息,使影像易于识别和分类 ...

2006

0.0

张锦水, 何春阳, 潘耀忠, 等. 基于SVM的多源信息复合的高空间分辨率遥感数据分类研究[J]. 遥感学报, 2006, 10(1): 49-57.

Zhang J

, He C

, Pan Y

, et al. The high spatial resolution RS image classification based on SVM method with the multi-source data[J]. Journal of Remote Sensing, 2006, 10(1): 49-57.

The RS miage shows a very promising perspective forurban land-coverand land-use classification,\nparticularlywith very high resolution(1―4m) satellite mi ages, while the traditional extractionmethods of the\nhigh spatial resolution mi age has the shortcomings of the low accuracy and classification efficiency. This paper\ndealswith the high spatial resolution mi age(IKONOS) classification based on the SVM method integrating the\ninformation of spectra,l texture and structure. And comparing to the results based onMaxmi um Likelihood and\nSVMmethodwith single-source data, this shows thatthe high spatial resolutionRS mi age classification based on\nSVMMethod with multi-source data can solve the mi age classification fragmentation which is based on the\nsingle-source data, spectrum, and has the good generalization ability with the high dmi ension vector. It has\nmore accuration than themaxmi um likelihoodmethod and SVM based on the single source data, adapts to the\nhigh spatial resolution RS Image classification.

遥感图像尤其是高空间分辨率（1―4m）遥感图像在土地利用和土地覆盖变化方面有着广阔的应用前景，传统高空间分辨率遥感图像信息提取方法存在精度和分类效率低的不足。本文提出的基于SVM的分类方法，复合光谱、纹理和结构信息等多源数据信息，对IKONOS高空间分辨率图像进行分类，并与最大似然法和单源数据（光谱）SVM分类结果进行定性和定量比较分析。研究结果表明，多源数据复合的SVM高空间分辨率遥感图像分类方法，能够有效解决单源数据信息图像分类效果破碎的问题；总精度达到68．38％，Kappa达到0．5993；对高维输入向量具有高的推广能力；比单源信息的SVM和最大似然方法图像分类精度更高，适合高空间分辨率遥感图像分类。

... 张锦水等^[4]基于支持向量机(support vector machine,SVM)的分类方法,利用复合光谱、纹理和结构信息等多特征数据对IKONOS高空间分辨率(简称#cod#x0201c ...

2007

0.0

黄昕, 张良培, 李平湘. 融合形状和光谱的高空间分辨率遥感影像分类[J]. 遥感学报, 2007, 11(2): 193-200.

Huang

, Zhang L

, Li P

Classification of high spatial resolution remotely sensed imagery based on the fusion of spectral and shape features[J]. Journal of Remote Sensing, 2007, 11(2): 193-200.

Classification of multispectral data with high spatial resolution is investigated.High spatial resolution remotely sensed(HSRRS) images with multispectral bands such as QuickBird and IKONOS provide a large amount of information,thus opening up avenues for new remote sensing applications.However,their availability poses challenges to image classification.Due to the complex spatial arrangement and spectral heterogeneity even within the same class,conventional spectral classification methods are grossly inadequate for HSRRS imagery.In order to overcome this inadequacy,spectral features must evidently be complemented by one or the other means.Shape is an important feature of high spatial resolution remotely sensed(HSRRS) imagery,and it is the manifestation of textures on such imagery.In this paper,a spatial feature index,pixel shape index(PSI),is proposed in order to describe the shape feature in a local area surrounding a pixel.PSI is a pixel based feature,which measures the gray similarity distance in every direction.As merely the shape feature is inadequate for classification of HSRRS imagery,the spectral bands are added to the input vectors of our classifier.And then a fast fusion algorithm which integrates both shape and spectral features using the support vector machine(SVM) has been developed to interpret the complex input vectors.To test the effectiveness of PSI,some spatial features extracted by wavelet transform(WT),gray level co-occurrence matrix(GLCM) are utilized for comparison.Experimental results demonstrate that PSI is capable of describing the shape features effectively and leads to more accurate classifications than other methods,meanwhile the classification algorithm fusing spectral and shape features by SVM is feasible.It is also found that spectral and shape features can complement each other and their integration can result in enhanced accuracy.

提出了一种像元形状指数及基于形状和光谱特征融合的高（空间）分辨率遥感影像分类方法。形状和光谱是遥感影像纹理的具体表现形式,尤其在高分辨率影像中地物细节得到充分表达,相邻像元的关系及其共同表征的形状特性成为分类的重要因素。本文用像元及其邻域的关系来描述其空间结构,同时为了更全面地利用影像特征,提出了基于支持向量机的形状和光谱融合分类方法。实验证明,该方法计算简便且能有效表达高分辨率影像的地物特征,提高分类精度。

... 黄昕等^[5]提出像元形状指数(pixel shape index,PSI),利用像元邻域的光谱相似性探测其空间形状特征,在高分遥感图像分类中具有一定的优势 ...

2003

0.0

... Shackelford等^[6]提出一种长度-宽度提取算法,用于提取光谱特性相似的一组像元的长度与宽度信息,并得出该算法适合对不同长/宽比对象进行分类的结论 ...

2012

0.0

... Han等^[7]提出的形状尺寸指数(shape-size index, SSI)则是基于频谱相似的同质区域中的中心像元及其相邻像元的特征信息提取的 ...

2010

0.0

薛丽霞, 李涛, 王佐成. 一种自适应的Canny边缘检测算法[J]. 计算机应用研究, 2010, 27(9): 3588-3590.

Xue L

, Li

, Wang Z

Adaptive Canny edge detection algorithm[J]. Application Research of Computers, 2010, 27(9): 3588-3590.

针对传统的Canny算子进行边缘检测时易丢失边缘细节的缺陷,提出了一种改进的Canny边缘检测算法.该算法采用自适应的滤波器对图像进行滤波,在滤除图像噪声的同时保留了更多的图像边缘信息;根据图像的灰度均值与方差均值自动计算Canny算子的高低阈值.实验结果表明,该算法在检测到更多边缘细节的同时也具有很强的自适应性.

... 2 实验方法与步骤本文的实验方法是基于光谱和形状特征,利用 SVM分类器进行分类: ①获得1景高分多光谱图像,用Canny滤波器^[8]提取边缘信息,得到模糊边缘图 ...

2011

0.0

... ④应用SVM分类器^[9]对图像进行基于光谱和形状特征的分类 ...

... SVM分类器的核函数为径向基函数(radial basis function, RBF),采用交叉验证获取最优参数C和g的值^[9] ...

2012

0.0

张华. 遥感数据可靠性分类方法研究[D]. 徐州: 中国矿业大学, 2012.

Zhang

Study on Reliable Classification Methods Based on Remotely Sensed Image[D].

Xuzhou: China University of Mining and Technology, 2012.

0.0

2007

0.0

2006

0.0