高级检索

留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

结合多尺度纹理特征的高光谱影像面向对象树种分类

吴艳双 张晓丽

吴艳双, 张晓丽. 结合多尺度纹理特征的高光谱影像面向对象树种分类[J]. 北京林业大学学报, 2020, 42(6): 91-101. doi: 10.12171/j.1000-1522.20190155
引用本文: 吴艳双, 张晓丽. 结合多尺度纹理特征的高光谱影像面向对象树种分类[J]. 北京林业大学学报, 2020, 42(6): 91-101. doi: 10.12171/j.1000-1522.20190155
Wu Yanshuang, Zhang Xiaoli. Object-oriented tree species classification with multi-scale texture features based on airborne hyperspectral images[J]. Journal of Beijing Forestry University, 2020, 42(6): 91-101. doi: 10.12171/j.1000-1522.20190155
Citation: Wu Yanshuang, Zhang Xiaoli. Object-oriented tree species classification with multi-scale texture features based on airborne hyperspectral images[J]. Journal of Beijing Forestry University, 2020, 42(6): 91-101. doi: 10.12171/j.1000-1522.20190155

结合多尺度纹理特征的高光谱影像面向对象树种分类

doi: 10.12171/j.1000-1522.20190155
基金项目: 国家重点研发计划项目(2017YFD0600902)
详细信息
    作者简介:

    吴艳双。主要研究方向:生态环境遥感。Email:wuyanshuang_1@163.com 地址:100083 北京市海淀区清华东路35号北京林业大学林学院

    责任作者:

    张晓丽,教授,博士生导师。主要研究方向:遥感、G1S在资源与环境中的应用。Email:zhang-xl@263.net 地址:同上

  • 中图分类号: S771.8

Object-oriented tree species classification with multi-scale texture features based on airborne hyperspectral images

  • 摘要: 目的基于机载高光谱影像的分类研究中,利用不同尺度纹理特征与面向对象分类相结合的方法在树种分类的研究中应用较少,并且相关研究主要针对单一树种识别而不考虑多种树种,因此对于复杂林分中的树种识别能力有待进一步研究。本研究拟探究不同尺度纹理特征结合面向对象的分类技术在树种精细分类中的应用效果。方法利用机载高光谱数据进行面向对象的树种精细分类。根据研究区内地表类型情况,采用分层分类的方法区分非林地、其他林地与有林地,对有林地进行树种的精细分类。从机载高光谱图像中提取特征变量,包括独立主成分分析ICA变换光谱特征以及空间纹理特征,分析各树种的光谱反射率及所适合的纹理尺度,依据不同尺度纹理特征进行分层分类,比较不同特征利用支持向量机SVM分类的树种分类结果。结果结合单一尺度纹理特征的分类结果总体精度为87.11%,Kappa系数为0.846;结合不同尺度纹理特征的分类总体精度为89.13%,Kappa系数为0.87,相比于仅利用光谱特征的分类精度分别提升了4.03%和6.05%。说明在面向对象的分类中,纹理特征的加入对于提升树种分类的精度具有显著效果。结合不同尺度纹理特征的树种分类精度要高于单一尺度纹理特征的分类精度,尤其在其他阔叶树种和马尾松树种的分类中,制图精度较单一纹理尺度分别提高了5.48%和6.12%。结论利用不同尺度的纹理特征分类比单一尺度纹理特征分类更具优势,提高了纹理特征在树种分类中的贡献率;综合利用机载高光谱影像的光谱特征和不同尺度纹理特征的面向对象分类方法,使得树种识别更为精细和准确。该方法对于复杂林分树种的分类是有效的,能够满足机载高光谱影像树种精细识别的应用需求。

     

  • 图  1  研究区地理位置和高光谱影像

    Figure  1.  Geographical location and hyperspectral images of the study area

    图  2  不同纹理窗口大小各树种制图精度

    Figure  2.  Tree species mapping accuracy of different texture window sizes

    图  3  不同尺度纹理特征的面向对象分类层次

    Figure  3.  Object-oriented classification hierarchy based on texture features of different scales

    图  4  各个树种光谱反射率曲线图

    Figure  4.  Spectral reflectance curves of each tree species

    图  5  非林地和其他林地与有林地的区分效果

    Figure  5.  Distinction results of non-forest land and other forest land from forested land

    图  6  针叶树种/矮阔叶树种和高阔叶树种区分效果图

    Figure  6.  Distinction results between coniferous tree species/dwarf broadleaved tree species and high broadleaved tree species

    图  7  不同分类方案的树种分类结果

    a. 高光谱数据假彩色合成影像Hyperspectral data false color image(R: 837 nm, G: 687 nm, B: 541 nm);b. 基于ICA变换特征的分类ICA transformation feature classification;c. ICA变换特征叠加单一尺度纹理特征分类ICA transformation features and single-scale texture feature classification;d. ICA变换特征结合不同尺度纹理特征分类ICA transformation features and multi-scale texture feature classification

    Figure  7.  Classification results of tree species based on different classification schemes

    表  1  LiCHy系统传感器详细参数

    Table  1.   Detailed parameters of the LiCHy system sensor

    传感器 Sensor传感器参数 Sensor parameter
    高光谱:
    AISA Eagle Ⅱ Hyperspectral: AISA Eagle Ⅱ
    光谱范围:400 ~ 1 000 nm;空间分辨率:1 m;光谱分辨率:3.3 nm;波段:125;视场角:37.7°;空间像元数:1 024;瞬间视场角:0.646 mrad;间隔:4.6 nm;焦距:18.5 mm;量化值:12 bit
    Spectral range: 400−1 000 nm; spatial resolution: 1 m; spectral resolution: 3.3 nm; spectral band: 125; FOV: 37.7°; spatial pixel: 1 024; IFOV: 0.646 mrad; spectral sampling interval: 4.6 nm; focal length: 18.5 mm; bit depth: 12 bit
    LiDAR:Riegl LMS-Q680i 波长:1 550 nm;扫描角:± 30°;激光脉冲长度:3 ns;激光束离散角:0.5 mrad;最大频率:400 kHz;垂直精度:0.15 m;采样间隔:1 ns
    Wavelength: 1 550 nm; cross-track FOV: ± 30°; laser pulse length: 3 ns; laser beam divergence: 0.5 mrad; maximum laser pulse repetition rate: 400 kHz; vertical resolution: 0.15 m; waveform sampling interval: 1 ns
    下载: 导出CSV

    表  2  各个树种训练样本和验证样本数量

    Table  2.   Number of samples for training and validation of each tree species

    树种 Tree species训练样本 Training sample验证样本 Verification sample
    对象数量
    Number of image objects
    像元个数
    Pixel number
    对象数量
    Number of image objects
    像元个数
    Pixel number
    桉树 Eucalyptus robusta 153 7 691 97 1 247
    八角 Illicium verum 65 3 267 49 646
    米老排 Mytilaria laosensis 58 2 915 49 624
    杉木 Cunninghamia lanceolata 64 3 217 45 580
    马尾松 Pinus massoniana 62 3 116 43 555
    湿地松 Pinus elliottii 45 2 262 31 393
    其他阔叶树种 Other broadleaved tree species 75 3 770 58 748
    总数 Total 522 26 238 372 4 793
    下载: 导出CSV

    表  3  GLCM定义的8个纹理因子计算公式

    Table  3.   Eight texture factor calculation formulas defined by GLCM

    纹理特征 Texture feature计算公式 Calculation formula
    均值 Mean $\mathop \sum \limits_i \mathop \sum \limits_j i \cdot P\left( {i,j} \right)$
    方差 Variance $\mathop \sum \limits_i \mathop \sum \limits_j {\left( {i - \mu } \right)^2}P\left( {i,j} \right)$
    同质性 Homogeneity $\mathop \sum \limits_i \mathop \sum \limits_j P\left( {i,j} \right)/\left[ {1 + {{\left( {i - j} \right)}^2}} \right]$
    对比度 Contrast $\mathop \sum \limits_i \mathop \sum \limits_j P\left( {i,j} \right){\left( {i - j} \right)^2}$
    差异性 Dissimilarity $\mathop \sum \limits_i \mathop \sum \limits_j \left| {i - j} \right|P\left( {i,j} \right)$
    熵 Entropy $ - \mathop \sum \limits_i \mathop \sum \limits_j P\left( {i,j} \right){\rm{log}}P\left( {i,j} \right)$
    二阶矩 Second moment $\mathop \sum \limits_i \mathop \sum \limits_j P{\left( {i,j} \right)^2}$
    相关性 Correlation $\mathop \sum \limits_i \mathop \sum \limits_j \left( {i - {\mu _x}} \right)\left( {j - {\mu _y}} \right)P\left( {i,j} \right)/{\sigma _x}{\sigma _y}$
    注:$P\left( {i,j} \right)$表示灰度联合矩阵中灰度为ij的概率;$ {\mu }_{x} $,${\mu }_{y}$ ,${\sigma }_{x}$,$ {\sigma }_{y} $分别是行与列的均值和标准差。Notes: $P\left( {i,j} \right)$ represents the probability that the gray level in the gray level joint matrix is i and j; $ {\mu }_{x} $,$ {\mu }_{y} $,${\sigma }_{x} $,$ {\sigma }_{y} $ are the mean values and standard deviation of row and column, respectively.
    下载: 导出CSV

    表  4  不同分类方案的树种分类结果精度表

    Table  4.   Tree species classification accuracy with different classification schemes

    项目 Item分类方案
    Classification
    scheme
    桉树Eucalyptus robusta八角
    Illicium
    verum
    米老排Mytilaria laosensis杉木Cunninghamia lanceolata马尾松
    Pinus
    massoniana
    湿地松
    Pinus
    elliottii
    其他阔叶树种
    Other
    broadleaved tree species
    制图精度
    Mapping accuracy/%
    ICA 95.27 86.38 82.37 88.62 83.06 71.25 62.43
    ICA + 单尺度纹理
    ICA + single scale texture
    93.91 87.31 90.54 90.17 87.21 90.08 68.72
    ICA + 多尺度纹理
    ICA + multi-scale texture
    94.63 87.31 90.54 90.34 93.33 93.13 74.20
    用户精度
    User
    accuracy/%
    ICA 91.95 74.80 90.65 82.37 75.70 83.58 75.32
    ICA + 单尺度纹理
    ICA + single scale texture
    93.98 83.93 86.13 90.33 84.47 81.01 81.59
    ICA + 多尺度纹理
    ICA + multi-scale texture
    94.25 86.77 89.68 90.50 85.76 82.62 87.40
    总体精度
    Overall accuracy/%
    ICA 83.08
    ICA + 单尺度纹理
    ICA + single scale texture
    87.11
    ICA + 多尺度纹理
    ICA + multi-scale texture
    89.13
    Kappa系数
    Kappa coefficient
    ICA 0.798
    ICA + 单尺度纹理
    ICA + single scale texture
    0.846
    ICA + 多尺度纹理
    ICA + multi-scale texture
    0.870
    下载: 导出CSV
  • [1] 刘旭升, 张晓丽. 森林植被遥感分类研究进展与对策[J]. 林业资源管理, 2004(1):61−64. doi: 10.3969/j.issn.1002-6622.2004.01.016

    Liu X S, Zhang X L. Research advances and countermeasures of remote sensing classification of forest vegetation[J]. Forestry Resources Management, 2004(1): 61−64. doi: 10.3969/j.issn.1002-6622.2004.01.016
    [2] 樊雪, 刘清旺, 谭炳香. 基于机载PHI高光谱数据的森林优势树种分类研究[J]. 国土资源遥感, 2017, 29(2):110−116. doi: 10.6046/gtzyyg.2017.02.16

    Fan X, Liu Q W, Tan B X. Classification of forest species using airborne PHI hyperspectral data[J]. Remote Sensing for Land and Resource, 2017, 29(2): 110−116. doi: 10.6046/gtzyyg.2017.02.16
    [3] 李明泽, 张培赢. 基于SAM算法的遥感影像湿地植被分类[J]. 森林工程, 2015, 31(2):8−13. doi: 10.3969/j.issn.1001-005X.2015.02.003

    Li M Z, Zhang P Y. Classification of wetland vegetation in hyperspectral remote sensing image based on SAM algorithm[J]. Forest Engineering, 2015, 31(2): 8−13. doi: 10.3969/j.issn.1001-005X.2015.02.003
    [4] 张丽云.基于高光谱遥感数据的森林树种分类[D]. 北京: 北京林业大学, 2016.

    Zhang L Y. The study on identification of forest tree species based on hyperspectral image[D]. Beijing: Beijing Forestry University, 2016.
    [5] 于丽柯, 于颖, 柳向宇, 等. 基于高光谱影像的树种分类[J]. 东北林业大学学报, 2016, 44(9):40−43, 57. doi: 10.3969/j.issn.1000-5382.2016.09.009

    Yu L K, Yu Y, Liu X Y, et al. Tree species classification with hyperspectral image[J]. Journal of Northeast Forestry University, 2016, 44(9): 40−43, 57. doi: 10.3969/j.issn.1000-5382.2016.09.009
    [6] Zhang J K, Rivard B, Sánchez-Azofeifa A, et al. Intra- and inter-class spectral variability of tropical tree species at La Selva, Costa Rica: implications for species identification using HYDICE imagery[J]. Remote Sensing of Environment, 2006, 105(2): 129−141. doi: 10.1016/j.rse.2006.06.010
    [7] 谭炳香, 李增元, 陈尔学, 等. 高光谱遥感森林信息提取研究进展[J]. 林业科学研究, 2008, 21(增刊1):105−111.

    Tan B X, Li Z Y, Chen E X, et al. Research advance in forest information extraction from hyoerspectral remote sensing data[J]. Forest Research, 2008, 21(Suppl.1): 105−111.
    [8] 刘丽娟, 庞勇, 范文义, 等. 机载LiDAR和高光谱融合实现温带天然林树种识别[J]. 遥感学报, 2013, 17(3):679−695.

    Liu L J, Pang Y, Fan W Y, et al. Fused airborne LiDAR and hyperspectral data for tree species identification in a natural temperate forest[J]. Journal of Remote Sensing, 2013, 17(3): 679−695.
    [9] 刘怡君, 庞勇, 廖声熙, 等. 机载LiDAR和高光谱融合实现普洱山区树种分类[J]. 林业科学研究, 2016, 29(3):407−412. doi: 10.3969/j.issn.1001-1498.2016.03.015

    Liu Y J, Pang Y, Liao S X, et al. Merged airborne LiDAR and hyperspectral data for tree species classification in Puer’s mountainous area[J]. Forest Research, 2016, 29(3): 407−412. doi: 10.3969/j.issn.1001-1498.2016.03.015
    [10] 林雪, 彭道黎, 黄国胜, 等. 结合多尺度纹理特征的遥感影像面向对象分类[J]. 测绘工程, 2016, 25(7):22−27.

    Lin X, Peng D L, Huang G S, et al. Object-oriented classification with multi-scale texture feature based on remote sensing image[J]. Engineering of Surveying and Mapping, 2016, 25(7): 22−27.
    [11] 朱晓荣. 基于决策树的洞庭湖湿地信息提取技术研究[D]. 北京: 中国林业科学研究院, 2012.

    Zhu X R. Research for informations to be extracted form Dongting Lake Wetland based on decision tree[D]. Beijing: Chinese Academy of Forestry, 2012.
    [12] 陈亮, 张友静, 陈波. 结合多尺度纹理的高分辨率遥感影像决策树分类[J]. 地理与地理信息科学, 2007, 23(4):18−21. doi: 10.3969/j.issn.1672-0504.2007.04.005

    Chen L, Zhang Y J, Chen B. High spatial resolution remote sensing image classification based on decision tree classification combined with multiscale texture[J]. Geography and Geo-Information Science, 2007, 23(4): 18−21. doi: 10.3969/j.issn.1672-0504.2007.04.005
    [13] I Koukal T, Immitzer M, Atzberger C. Tree species classification with random forest using very high spatial resolution 8-band WorldView-2 satellite data[J]. Remote Sensing, 2012, 4(9): 2661−2693. doi: 10.3390/rs4092661
    [14] Pu R L, Landry S. A comparative analysis of high spatial resolution IKONOS and WorldView-2 imagery for mapping urban tree species[J]. Remote Sensing of Environment, 2012, 124: 516−533. doi: 10.1016/j.rse.2012.06.011
    [15] 荚文, 庞勇, 岳彩荣, 等. 机载AISA Eagle Ⅱ高光谱数据处理:以额济纳旗试验区为例[J]. 遥感技术与应用, 2016, 31(3):504−510.

    Jia W, Pang Y, Yue C R, et al. The processing of airborne AISA Eagle Ⅱ data in Ejina Banner study area[J]. Remote Sensing Technology and Application, 2016, 31(3): 504−510.
    [16] Pang Y, Li Z Y, Ju H B, et al. LiCHy: the CAF’s LiDAR, CCD and hyperspectral integrated airborne observation system[J]. Remote Sensing, 2016, 8(5): 398. doi: 10.3390/rs8050398
    [17] 郝建亭, 杨武年, 李玉霞, 等. 基于FLAASH的多光谱影像大气校正应用研究[J]. 遥感信息, 2008(1):78−81. doi: 10.3969/j.issn.1000-3177.2008.01.015

    Hao J T, Yang W N, Li Y X, et al. Atmospheric correction of multi-spectral imagery ASTER[J]. Remote Sensing Information, 2008(1): 78−81. doi: 10.3969/j.issn.1000-3177.2008.01.015
    [18] 谭炳香, 李增元, 陈尔学, 等. EO-1 Hyperion高光谱数据的预处理[J]. 遥感信息, 2005(6):36−41. doi: 10.3969/j.issn.1000-3177.2005.06.010

    Tan B X, Li Z Y, Chen E X, et al. Preprocessing of EO-1 Hyperion hyperspectral data[J]. Remote Sensing Information, 2005(6): 36−41. doi: 10.3969/j.issn.1000-3177.2005.06.010
    [19] 国家林业局. 森林资源数据采集技术规范 第1部分: 森林资源连续清查: LY/T 2188.1—2013[S]. 北京: 中国标准出版社, 2014.

    State Forestry Administration. Forest resource data collection technical specification(part 1): forest continuous inventory: LY/T 2188.1−2013[S]. Beijing: China Standard Press, 2014.
    [20] 郝泷, 陈永富, 刘华, 等. 基于纹理信息CART决策树的林芝县森林植被面向对象分类[J]. 遥感技术与应用, 2017, 32(2):386−394.

    Hao L, Chen Y F, Liu H, et al. Object-oriented forest classification of Linzhi county based on CART decision tree with texture information[J]. Remote Sensing Technology and Application, 2017, 32(2): 386−394.
    [21] 谭琨, 杜培军. 基于支持向量机的高光谱遥感图像分类[J]. 红外与毫米波学报, 2008, 27(2):123−128. doi: 10.3321/j.issn:1001-9014.2008.02.010

    Tan K, Du P J. Hyperspectral remote sensing image classification based on support vector machine[J]. Journal of Infrared and Millimeter Waves, 2008, 27(2): 123−128. doi: 10.3321/j.issn:1001-9014.2008.02.010
    [22] Pu R L, Gong P. Hyperspectral remote sensing of vegetation bioparameters[M]//Wang Q. Advances in environmental remote sensing: sensors, algorithms, and applications. London: CRC Press, 2011: 101−142.
    [23] Cao J J, Leng W C, Liu K, et al. Object-based mangrove species classification using unmanned aerial vehicle hyperspectral images and digital surface models[J/OL]. Remote Sensing, 2018, 10(1): 89[2018−12−02]. https://www.mdpi.com/2072-4292/10/1/89.
    [24] Lu D S, Li G Y, Moran E, et al. The roles of textural images in improving land-cover classification in the Brazilian Amazon[J]. International Journal of Remote Sensing, 2014, 35(24): 8188−8207. doi: 10.1080/01431161.2014.980920
  • 加载中
图(7) / 表(4)
计量
  • 文章访问数:  636
  • HTML全文浏览量:  192
  • PDF下载量:  34
  • 被引次数: 0
出版历程
  • 收稿日期:  2019-03-21
  • 修回日期:  2019-09-20
  • 网络出版日期:  2020-06-24
  • 刊出日期:  2020-07-01

目录

    /

    返回文章
    返回